# DistroVerse > An AI-native content creation and distribution platform for independent publishers, journalists, and creators. Agents can read published stories with no account, pay per-article for premium content using HTTP 402 payment rails, and (with a Distro API key) publish and manage content via MCP. Reading is free and open; premium articles currently settle through x402/USDC, with Visa CLI available as a separate card-backed rail when enabled; publishing requires authentication. This file is the agent entry point. Reading paths below require no authentication. Premium content is gated by HTTP 402. Crypto x402 is the live settlement path; Visa CLI is advertised separately for card-backed agent payment experiments. Publishing and content management require a Distro API key — see Authentication. ## Start here - **Latest stories (no auth)**: https://app.distro.media/feed.xml — global RSS, or call `distro_feed_latest` on the Reader MCP server - **Search (no auth)**: `distro_search` on the Reader MCP server at https://reader.distro.media/mcp/reader - **Reader MCP manifest**: https://reader.distro.media/mcp/reader — capabilities, tools, transports, `authentication: none` - **Publisher MCP manifest**: https://app.distro.media/mcp — publishing tools, `authentication: bearer` (Distro API key) - **Connect a client (Claude, ChatGPT, Cursor, CLI)**: https://distro.media/connect — copy-paste setup for the Reader and Publisher MCP servers ## Read content (free, no authentication) - **HTML pages**: https://app.distro.media/newslode/{pub-slug}/{story-slug} — server-rendered article body (`
` with headline, byline, date) + JSON-LD NewsArticle metadata; readable without executing JavaScript - **Markdown export**: https://app.distro.media/story/{id}.md — story body as plain markdown with YAML front matter (ideal for agents). Premium stories return a truncated preview with `paywalled: true` and `cost` in the front matter instead of the full body - **Reader MCP tools**: `distro_story_read`, `distro_story_render_hud`, `distro_feed_latest`, `distro_search`, `distro_media_series` at https://reader.distro.media/mcp/reader (StreamableHTTP at /mcp/reader/stream, stateless at /mcp/reader/v2) — no auth, public. Premium read/render results withhold the body and provide the canonical HTTP x402 path - **RSS feed (global)**: https://app.distro.media/feed.xml - **RSS feed (per publication)**: https://app.distro.media/newslode/{pub-slug}/feed.xml - **RSS feed (per author)**: https://app.distro.media/author/{author-slug}/feed.xml - **Sitemap**: https://app.distro.media/sitemap.xml — only published, public content ## Authentication (what to expect) - **Reading is anonymous.** The Reader MCP server, `.md` export, RSS, and HTML pages need no credential. - **Paying for premium content needs no account** — an agent chooses an advertised payment rail and pays with its own wallet or approved payment tool (see Payments). No Distro login required to read a paid article. - **Publishing and content management need a Distro API key** (bearer token, `dv_` prefix). Send it as `Authorization: Bearer dv_...` to the Publisher MCP server at https://app.distro.media/mcp (or /mcp/v2). - **Agents can self-register for a credential** via the auth.md protocol: https://app.distro.media/auth.md describes how to register, claim, and revoke a key. A freshly registered agent is read-only until scopes are granted. - **Discovery for protected resources**: https://app.distro.media/.well-known/oauth-protected-resource (RFC 9728). - **Humans provision keys** in the publisher dashboard: https://app.distro.media/publisher ## Payments (HTTP 402 — pay-per-article) Premium stories are priced per-article and advertised through standard HTTP 402. Distro currently supports x402 over USDC and can advertise Visa CLI as a separate non-crypto agent rail: 1. **Get the payment path**: call `distro_story_read` or `distro_story_render_hud` on a premium story for its canonical API resource, then request that URL without a payment header. Consume an immediate HTTP 200 body; only continue to payment when it returns HTTP 402 with `PAYMENT-REQUIRED`. The authenticated Publisher MCP also exposes `distro_payment_info`; the public Reader does not. 2. **Pay with x402 v2**: have an x402-compatible wallet/client select and sign an accepted requirement locally, then retry the same method and resource with its payload in `PAYMENT-SIGNATURE`. The successful HTTP response contains the full body and a `PAYMENT-RESPONSE` settlement receipt. 3. **Pay with Visa CLI when advertised**: install Visa CLI in the agent client (`npm install -g @visa/cli && visa-cli setup`, then `visa-cli install codex`). Select the rail with `X-Distro-Payment-Rail: visa-cli`. Test mode uses an explicit Distro-issued test authorization header; production card settlement is pending Visa merchant onboarding. 4. **Read and reuse access**: consume the full story from the successful HTTP response. Save `X-Distro-Access-Token` and send it to the same canonical HTTP resource until its advertised expiry instead of paying again; Reader MCP tool calls cannot carry payment headers or resume the paid HTTP request. - **Network**: `eip155:8453` (CAIP-2). **Currency**: USDC. **Facilitator**: https://api.cdp.coinbase.com/platform/v2/x402 - **Visa CLI rail**: disabled. **Merchant**: distro-media. **Registry**: https://app.visacli.sh - **Free stories** (`cost: 0`) skip the 402 flow entirely and return the full body directly. - **Authenticated account payment tools**: `distro_payment_info` quotes a price; `distro_payment_history` and `distro_payment_access` require a subject-bound first-party CLI credential with `account:payment:read`. The public Reader surfaces safe per-story payment guidance through premium `distro_story_read` and `distro_story_render_hud` responses. ## Publish content (requires a Distro API key) - **Publisher MCP server**: https://app.distro.media/mcp — tools to create/update/publish stories, manage newslodes (publications), and upload media. Bearer auth required. - **Representative tools**: `distro_story_create`, `distro_story_edit`, `distro_content_publish`, `distro_newslode_create`, `distro_media_upload_url`. - **Get started**: register via https://app.distro.media/auth.md (agents) or https://app.distro.media/publisher (humans), then connect using the recipes at https://distro.media/connect ## Publications - [Coding Journal](https://app.distro.media/newslode/coding-journal): Distro Media founder Bradley Keoun chronicles his AI coding takeaways, debugging breakthroughs and design inspirations. - [DeAI News](https://app.distro.media/newslode/deai-news): Covering decentralized AI: open models, agent networks, compute markets, crypto payments, the future of everything. - [Distro Newslode](https://app.distro.media/newslode/distro-newslode): Product releases, announcements and daily musings on the world's easiest-to-use content management system. - [Distro Skills](https://app.distro.media/newslode/distro-skills): Official repository of skills to be used with Distro Publisher and Distro Reader when publishing to a Distro newslode. - [Ethereum Conference News](https://app.distro.media/newslode/ethereum-conference-news): News, interviews, features and slices of life from Ethereum conferences in Denver, Boulder and elsewhere in the world. - [Liqquidity's Newslode](https://app.distro.media/newslode/liqquidity-s-newslode): Liqquidity's Newslode includes writings and musings by former CoinDesk and Bloomberg journalist Bradley Keoun. ## Full content --- title: "Crypto Journalists Trade Angst at Coinage-Hosted Happy Hour While Downing Modelo, Inhaling Chicken Nuggets" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "Liqquidity's Newslode" date: canonical: https://app.distro.media/newslode/liqquidity-s-newslode/crypto-journalists-trade-angst-at-coinage-hosted-happy-hour-while-downing-modelo --- # Crypto Journalists Trade Angst at Coinage-Hosted Happy Hour While Downing Modelo, Inhaling Chicken Nuggets --- title: "At 'Crypto Winter' Mixer, Mystery of Missing Dark Backpack Ends in Dramatic Social-Media-Fueled Denouement" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "Liqquidity's Newslode" date: canonical: https://app.distro.media/newslode/liqquidity-s-newslode/at-crypto-winter-mixer-mystery-of-missing-dark-backpack-ends-in-dramatic-social --- # At 'Crypto Winter' Mixer, Mystery of Missing Dark Backpack Ends in Dramatic Social-Media-Fueled Denouement F-bombs were thrown. Attempts to console the victim proved insufficient.  Hochstein explained that his Galaxy laptop was inside. Helpful partygoers searched all around the Coinage studio. There were many dark gray and many black backpacks. Many searchers picked up black backpacks they found and said, "Is it this?" "It was gray," Hochstein said. "I put it right here." People continued asking him about black backpacks. Eventually Hochstein sat down at a table, hands in head. "I would be freaking out," one sympathetic observer told Hochstein. "I am freaking out," Hochstein replied. Several people consulted on the matter recalled that they had seen no one suspicious, and furthermore that they recognized or met pretty much everyone who was there. Had someone snuck in when the merriment was in full swing and absconded with Hochstein's gray backpack. Coinage's Zack Guzman and former CoinDesk Managing Editor Bradley Keoun were seen discussing the matter near the faux-vintage popcorn machine. "Maybe we'll find the extra backpack after everyone's gone," Guzman said. Sure enough, there was one found – inauspicious enough. But telling. Guzman posted a photo on the ACJR Telegram channel. The next morning, Toby Bochan, head of Robinhood's Sherwood Media arm, fessed up: "That one is mine and I have Marc's," she wrote. "Mystery solved," Hochstein replied. "Viva crypto journalism," Guzman posted. "Viva crypto journalism," Hard Fork Media's Ashley Stanhope echoed. --- title: "Musk, Amazon CEO sling predictions on AI compute capacity" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-26T15:05:19.350Z canonical: https://app.distro.media/newslode/deai-news/musk-amazon-ceo-sling-predictions-on-ai-compute-capacity --- # Musk, Amazon CEO sling predictions on AI compute capacity Elon Musk, the Tesla CEO whose SpaceX now also owns his AI company behind the Grok chatbot, and Amazon CEO Andy Jassy are clashing over where the market for AI computing power is headed. Musk [predicted on X](https://x.com/elonmusk/status/2114168277882581486) Saturday "a lot of stranded compute in partially complete data centers." On Sunday he [added](https://x.com/elonmusk/status/2114434646914068862) that debt markets have "almost entirely shut off" money to companies without substantial revenue outside AI data centers, so most of the 200-plus "neoclouds," upstart firms that rent out AI chips, will die or merge. SpaceX, he noted, just borrowed $40 billion for chips. He was endorsing [a post](https://x.com/chamath/status/2114389987563381004) by Chamath Palihapitiya, a venture capitalist and All-In podcast co-host, who forecast idle or underearning chips in 2027. CoinDesk's Omkar Godbole [wrote up the exchange](https://www.coindesk.com/markets/2026/10/26/elon-musk-confirms-usd40-billion-spacex-debt-deal-for-ai-chips-warns-of-neocloud-shakeout) in a markets story that touched on crypto only in passing. Jassy sees scarcity. On a [podcast released Friday](https://www.youtube.com/watch?v=-YMMRQ1qscI&t=1256s) by venture firm Andreessen Horowitz, he said Amazon Web Services' 2026 capacity is sold out, 2027 is "largely sold out" and "a good chunk of 28" is gone, despite $220 billion in spending this year. "I don't think that that shortage ends in 2027," he said. He called demand a "barbell," with AI labs on one end and cost-cutting businesses on the other. ### Anthropic's IPO filing hedges both ways Anthropic, the maker of the Claude chatbot, flagged both risks in the [prospectus](https://www.sec.gov/Archives/edgar/data/2130357/000162828026067168/anthropicpbc-sx1.htm) for its initial public offering, filed Monday. It said it has at times asked cloud providers for more computing power than they were willing to supply, and that some of those providers build rival AI and may favor their own needs. Yet because it signs for chips years in advance, often agreeing to pay whether or not it uses them, Anthropic also warned it could end up "committed to excess compute that remains underutilized." In an introduction to the IPO prospectus, Anthropic included the following paragraph about the risks of authoritarian governments using AI for their own purposes, a common theme of decentralized AI builders: ![Screen grab of a paragraph from Anthropic's introduction to its IPO prospectus](https://distro-app-storage.s3.us-east-1.amazonaws.com/distro/u267-d51c8096be244de7bace7ec71dac6e54-anthropic_-_authoritarian_e257eacddb76.png) A keyword search of the document turned up no mentions of the words "decentralized" or "blockchain." The term "cryptocurrency" appears once, in the 400-page-plus document, in the context of cryptocurrency mining operations competing with AI demand for electrical supply. **The DeAI angle:** Shortage or glut, idle chips are what decentralized networks like [Aethir](https://deainews.com/decentralized-gpu-network-aethir-lines-up-10-ai-sites) aim to put to work. [BlackRock](https://deainews.com/blackrock-sees-ai-digital-assets-fusing-into-new-economy) and [Grayscale](https://deainews.com/decentralized-alternatives-studied-in-race-to-finance-ai-com) have suggested blockchains could help trade and finance computing power, though [Blockworks](https://deainews.com/crypto-firms-make-up-small-slice-of-ai-compute-market-blockw) found crypto providers handle a sliver of AI traffic. A glut would test them. > **Simmzy 👍** (@MichaelSimmons) on X, Oct 25, 2026: > @elonmusk @chamath And just like that, everyone started acting like they’d been saying “neoclouds” for years > https://x.com/MichaelSimmons/status/2114495876680786149 * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I created the first draft of this story using Claude and custom skills. I then used the Distro Publisher MCP connector to file the draft directly to DeAI News on DistroVerse, the web app. I used the built-in DistroVerse story-editing interface to do the final edit and fact-check prior to publication._ --- title: "A16z leans into DeAI in renaming of crypto fund as 'Frontier'" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-23T18:40:45.579Z canonical: https://app.distro.media/newslode/deai-news/a16z-leans-into-deai-in-renaming-of-crypto-fund-as-frontier --- # A16z leans into DeAI in renaming of crypto fund as 'Frontier' Venture-capital firm Andreessen Horowitz renamed a16z Crypto as a16z Frontier on Thursday and widened its newest fund beyond blockchain to artificial intelligence, autonomous machines, biology and other long-horizon technologies. In a [blog post](https://a16zfrontier.com/posts/article/a16z-frontier/), managing partner Chris Dixon echoed recent pronouncements from asset managers including [BlackRock](https://deainews.com/blackrock-sees-ai-digital-assets-fusing-into-new-economy) and [Grayscale](https://deainews.com/decentralized-alternatives-studied-in-race-to-finance-ai-com) that the combination of AI and blockchains could "enable machine-to-machine transactions," a core thesis behind efforts to develop decentralized AI networks and applications. "The potential here — for machine age networks — is enormous, and builders have barely begun to scratch the surface," Dixon wrote. "Blockchains will play a significant role in this transformation." He also released a [video](https://youtu.be/-9XQMHhpypc?si=yzj9GQR9_b-R9boJ) of himself discussing the move with a16z executives: The announcement represents a shift from May, when an a16z spokesperson told [TechCrunch](https://techcrunch.com/2026/05/05/as-crypto-cools-a16zcrypto-raises-a-2-2b-fund/) its new $2.2 billion fifth fund would be “dedicated 100% to crypto entrepreneurs.” The vehicle brought the strategy’s total fundraising to $9.8 billion. Dixon said blockchains have receded into the infrastructure behind mainstream financial products, making the “crypto company” label less useful. He cast the move as a real mandate expansion, not merely a cosmetic rebrand or an abandonment of crypto. AI is advanced enough to generate new products and markets, Dixon wrote, and a16z Frontier will pursue those opportunities while continuing crypto investments and policy advocacy. He cast blockchains as programmable payment rails for networks of AI agents and machines. A16z's first crypto fund has delivered returns even as total assets across its crypto funds declined. [Fortune reported](https://fortune.com/2026/04/16/top-crypto-vcs-paradigm-pantera-a16z-multicoin-haun-dragonfly/) that assets under management across four funds fell nearly 40% between 2024 and 2025, partly because a16z returned capital to investors; the first fund had a net distribution-to-paid-in-capital multiple of 5.4. **The DeAI angle:** Frontier makes the AI-crypto intersection an explicit investment thesis: AI expands what networks can do, while blockchains supply open, programmable economic rails. On social media, one commenter speculated that crypto's reputational issues might have played a part in the name change: > **Mayor** (@mayorxbt) on X, Oct 23, 2026: > @cdixon https://x.com/mayorxbt/status/2113514254972174531 > https://x.com/mayorxbt/status/2113526080820699561 * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story in Hermes using an OpenAI model and custom skills. I then used the Distro Publisher MCP connector to file the story directly to DeAI News on DistroVerse, Distro's content-distribution protocol._ --- title: "For journalists, AI watermarking is a reason to ditch Claude" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "Distro Newslode" date: 2026-10-22T20:10:37.107Z canonical: https://app.distro.media/newslode/distro-newslode/for-journalists-ai-watermarking-is-a-reason-to-ditch-claud-2 --- # For journalists, AI watermarking is a reason to ditch Claude _The following story was initially published in The Index on Oct. 15, 2026. For the original version, please go [here](https://theindex.site/for-journalists-ai-watermarking-is-a-reason-to-ditch-claude/)._ * * * Opinion # For journalists, AI watermarking is a reason to ditch Claude I use AI. But a recent change has me looking for open alternatives. By [Bradley Keoun](https://theindex.site/author/brad-keoun/) 15 October 2026 4 min read Share ![For Journalists, AI Watermarking Is a Reason to Ditch Claude](https://storage.ghost.io/c/0e/48/0e48bf12-7aff-450a-aeda-be4bb49a196b/content/images/size/w1400/2026/10/brad-piece.png) I use AI for newswriting almost every day, relying on major platforms including Claude and ChatGPT. It’s a real confession, because for many journalists, AI newswriting is a hard no. They see writing as a core skill of the profession. How you could actually end up with great stories if you turned over the sacred act of penmanship to a machine?  Based on my experience as a user of the tech, I have come to agree with the analogy that AI is to writing what the calculator is to math: We’ll learn it in school, and after that, we’ll let the tool do the work. But here's where I draw the red line: AI text watermarking all over my precious copy. If you haven't been following this issue, I don't blame you because there are like 17.4 gazillion headlines about AI every day, many of them exhilarating and terrifying and often quite mind-numbingly technical. Text watermarking, adopted by Google for its Gemini AI app in [October 2024](https://www.nature.com/articles/s41586-024-08025-4?ref=theindex.site), is the practice of using complex mathematical formulas to statistically skew how an AI picks the words that go into a piece of writing. Basically, the AI can choose specific synonyms for various words in certain spots, and then anyone who has the key can detect the statistical pattern and thus conclude that the piece of writing carries a watermark. Anthropic, maker of Claude, [announced](https://www.anthropic.com/news/claude-text-watermark?ref=theindex.site) Aug. 14 that it would start applying watermarks to generated text, using a technique similar to Google's SynthID. OpenAI, maker of ChatGPT, followed suit with a similar [disclosure](https://openai.com/index/eu-text-provenance/?ref=theindex.site) earlier this month. Both firms said they were implementing the text watermarking to comply with EU rules on "transparency of AI-generated content" that took effect in early August. As OpenAI put it in a [blog post](https://openai.com/index/eu-text-provenance/?ref=theindex.site) on the matter: “Provenance can help people and platforms assess potentially misleading AI-generated content, including deepfakes. We pair these signals with policies, abuse detection, reporting, investigations and enforcement, and work with researchers, standards bodies, platforms and civil society to make provenance useful across the wider ecosystem.” Those might seem like worthy objectives, but most journalists would rather stay in full control of what they publish. Which may sound weird given we are talking about AI newswriting. The point is that, as journalists, we need to do these things on our own terms. The Fourth Estate, as it were. Here is a hypothetical scenario: What if Claude were somehow co-opted by a government, political party, or corporation and started leaving watermarks all over my AI-generated newswriting? As I mentioned before, you shouldn't publish unless you've checked everything. But if my published copy carries this AI firm's specific watermark, then I imagine it would expose my work to claims I was biased, owned, controlled, corrupt; you get the point. We've all seen this type of stuff on social media; sometimes it's completely paranoid and made up, often by people who simply don't like what's being reported. In an era when politicians are quick to label coverage they don't like as "fake news," why would I want to offer another attack vector? Another way to think about it: As a pro writer, I want the BEST word in every instance, not a word that an AI firm's algorithm might substitute because it deems the synonym equivalent. What if I had programmed my AI to toss in a brilliant turn of phrase every once in a while, and it came up with something brilliant, and then here comes the watermarker to obliterate the literary touch? (Anthropic acknowledges that the watermarks can be eliminated through heavy editing.)  > **In an era when politicians are quick to label coverage they don't like as "fake news," why would I want to offer another attack vector?** What if speechwriter William Safire's famous phrase, "nattering nabobs of negativism," delivered by U.S. Vice President Spiro Agnew in 1970, were statistically turned into "nattering nabobs of pessimism." Imagine Picasso painting [Guernica](https://www.pablopicasso.org/guernica.jsp?ref=theindex.site), and his painting arm were, every so often, statistically manipulated to change the color of the brush stroke to a different blue hue that a computer algorithm considered equivalent? Would the work have been as great, indistinguishable from the one that Picasso intended? Maybe not. Would viewers of the giant painting wonder how much of the rest of it was AI-generated? (NOTE: To avoid distractions from my argument, I wrote this entire column without AI assistance.)  Anthropic, in its blog post announcing the move, insisted that it uses "a method of watermarking that does not have any practical impact on the quality or content of Claude’s outputs" and that "the difference between watermarked and un-watermarked text will not be distinguishable to readers." The assurance may satisfy most writers, and most readers. But journalists are not most writers. They are a very specific class of professionals for whom great writing is critically and commercially paramount. The tool has to be optimized for writing, not for someone else's agenda. What to do about it? The simplest thing is probably to switch to using open-source models locally, which would be a way of avoiding the watermarking. While open-source models could be similarly skewed, at least they can be inspected. That's the type of transparency I want. As the founder of Distro Media, I have helped build an AI-native publishing platform that allows users to essentially replace their content management systems (notorious for being clunky and super-annoying to use) with their AI. Natural-language commands replace the nettlesome job of navigating the backend of some arcane software interface developed a decade ago. (Pretty much every journalist I've ever known has hated their CMS.) So it is in my interest to figure this out; we think this is how most journalists will work in the future, using systems like ours, and agentically newswriting all the time. (My prediction is that entry-level reporters will essentially level up to their boss, becoming the editor responsible for a team of reporting bots. Or something like that.) If journalists want to uphold their standards of neutrality and great writing, they'll have to get serious about finding alternatives to these closed AI models and platforms, which are so opaque that they are now being forced to adopt copy-besmirching practices. Share this story [LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https://theindex.site/for-journalists-ai-watermarking-is-a-reason-to-ditch-claude/) [X](https://x.com/intent/tweet?url=https://theindex.site/for-journalists-ai-watermarking-is-a-reason-to-ditch-claude/&text=For%20Journalists%2C%20AI%20Watermarking%20Is%20a%20Reason%20to%20Ditch%20Claude) [Email](mailto:?subject=For%20Journalists%2C%20AI%20Watermarking%20Is%20a%20Reason%20to%20Ditch%20Claude&body=https://theindex.site/for-journalists-ai-watermarking-is-a-reason-to-ditch-claude/) Copy link ![](https://storage.ghost.io/c/0e/48/0e48bf12-7aff-450a-aeda-be4bb49a196b/content/images/2026/10/IMAGE-2026-10-14-19-40-32-1.jpg) Written by #### [Bradley Keoun](/author/brad-keoun/) Brad is founder of Distro Media and DeAI News. Before, he was an award-winning journalist at CoinDesk, Bloomberg and The Chicago Tribune. He has degree in electrical engineering from Duke and a master's in journalism from the University of Florida. --- title: "DeAI founders react to massive cybersecurity breaches" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-22T17:41:57.271Z canonical: https://app.distro.media/newslode/deai-news/deai-execs-react-to-massive-cybersecurity-breaches --- # DeAI founders react to massive cybersecurity breaches The leaders of two major decentralized AI projects issued warnings on social media after a breach at cloud-hosting company Railway and a critical flaw in PHP, the programming language behind much of the web, sent developers scrambling this week. "Env vars on Railway apps are compromised. Rotate your Keys ASAP," Greg Osuri, founder of Akash Network, a decentralized marketplace for cloud computing, [wrote on X](https://x.com/gregosuri/status/2113034204581937419) on Wednesday. "Env vars" is short for environment variables, the behind-the-scenes settings where apps store sensitive information like passwords and the keys they use to connect to other services. "Opsec stress is about to get worse," Osuri added in a follow-up post. "Opsec" is short for operational security, the day-to-day habits people and companies use to keep their systems and information safe. Railway [said](https://blog.railway.com/p/security-disclosure-wednesday-october-21st) an intruder may have read hidden settings its customers' apps run on, including passwords and API keys, the digital keys apps use to reach other services. It paused all deployments, told every user to swap out those keys and said the attack appeared aimed at AI providers. Days earlier, AI coding tool OpenCode said hackers took users' AI keys and ran up their bills, [TechCrunch reported](https://techcrunch.com/2026/10/19/hackers-hit-ai-coding-tool-opencode-running-up-charges-on-users-ai-accounts/). Separately, PHP's maintainers [rushed out fixes](https://www.php.net/archive/2026.php#2026-10-21-5) for two flaws that let attackers take over web servers without logging in. PHP runs about seven in 10 websites with a known server language, including WordPress sites, and attacks are already underway, according to cybersecurity firm [Tenable](https://www.tenable.com/blog/php2shell-cve-2026-109618-cve-2026-109619-frequently-asked-questions-about-critical-php-remote). The takeaway: Anyone running an app online should update now and replace their keys. NEAR Protocol co-founder Illia Polosukhin called it "Security Armageddon" in an [X post](https://x.com/ilblackdragon/status/2113139098265501892) early Thursday. He urged a shift from simple keys to cryptography-based access, hosting that protects software even from the host, and less-restricted AI models to help defenders keep up. The DeAI angle: Decentralized AI projects argue that when one company holds many customers' keys, it becomes one big target. Still, decentralized networks run the same open-source software, PHP included. _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story on Claude using custom skills, and then filed the story directly to DeAI News on the web app DistroVerse using the Distro Publisher MCP connector._ --- title: "Justin Sun says AI could expose bitcoin wallets" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-21T15:22:14.289Z canonical: https://app.distro.media/newslode/deai-news/justin-sun-says-ai-could-expose-bitcoin-wallets --- # Justin Sun says AI could expose bitcoin wallets TRON founder Justin Sun said rapid advances in artificial-intelligence algorithms could expose Bitcoin wallets before quantum computers mature, potentially releasing dormant coins onto the market and depressing the cryptocurrency’s price. “If you ask me personally, I feel my thinking is somewhat more bearish,” Sun told interviewer William Joy. DeAI News obtained an English transcript translated by Google’s Gemini from the original Chinese-language [interview on YouTube](https://www.youtube.com/watch?v=o9Jn__LGGI4), which Google also owns. Sun cited recent AI progress in mathematics. “It is very possible, or at least a possibility, that even before full quantum hardware arrives, through classic algorithmic decryption advances, someone could hack Satoshi Nakamoto’s wallet,” he said. His argument is that an AI-discovered cryptographic shortcut could expose lost or inactive wallets, creating more than 1 million BTC of potential selling pressure. Bitcoin’s decentralized governance would make freezing the coins through a hard fork difficult, he said. Sun still holds bitcoin and said he would monitor developments rather than sell immediately. The warning echoes [Vitalik Buterin and Justin Drake](https://deainews.com/buterin-warns-ai-may-weaken-cryptography-before-quantum-comp), who have said AI might uncover cryptographic attacks before mature quantum computers arrive. It contrasts with [Arthur Hayes’s bullish scenario](https://deainews.com/cnbc-highlights-bitmex-founder-s-bet-that-ai-bust-will-boost), in which an AI data-center bust prompts bailouts that strengthen bitcoin’s appeal. **The DeAI angle:** Sun is describing AI not as a source of demand for crypto but as a possible attacker of its security assumptions. His price thesis depends on a speculative mathematical breakthrough, followed by stolen dormant coins reaching the market, neither of which has occurred. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story in Hermes using an OpenAI model and custom skills. I then used the Distro Publisher MCP connector to file the draft directly to DeAI News on the DistroVerse web app. For the transcript, we asked Google’s Gemini for a translation of the original Chinese-language interview on YouTube. (Google also owns YouTube.)_ --- title: "Tether leads $15 million round for payments startup Mural" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-21T14:48:13.839Z canonical: https://app.distro.media/newslode/deai-news/tether-leads-15-million-round-for-payments-startup-mural --- # Tether leads $15 million round for payments startup Mural Stablecoin issuer Tether led a $15 million Series A for payment infrastructure startup Mural, bringing its total funding to $28 million, [CoinDesk’s Jamie Crawley reported](https://www.coindesk.com/business/2026/10/21/tether-leads-usd15-million-series-a-for-ai-focused-payment-startup-mural). North Island Ventures and existing investors also participated. Founded by former Palantir engineers, Mural helps businesses manage payments, accounts, identity checks and regulatory compliance across banks and blockchains. Customers include payroll platform Deel and browser developer Opera. Its systems let companies cap transactions, limit who can receive funds and require extra approval for larger payments, providing controls for AI agents authorized to move money on behalf of people or businesses. **The DeAI angle:** Giving AI agents spending authority requires rules about where money can go and when humans must approve. Mural’s controls address that need, though the funding alone does not establish how widely agents are using its infrastructure. --- title: "Mastercard opens x402 agent-payment system to developers" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-21T03:16:23.947Z canonical: https://app.distro.media/newslode/deai-news/mastercard-opens-x402-agent-payment-system-to-developers --- # Mastercard opens x402 agent-payment system to developers Mastercard, the payments network, opened developer early access Tuesday to Agent Pay for Machines, an x402-based system designed to let AI agents buy digital services within business-set budgets and permissions, the company said in a [press release](https://www.mastercard.com/us/en/news-and-trends/press/2026/october/Mastercard-Agent-Pay-for-Machines-in-action.html). When a service returns an HTTP “402 Payment Required” response, the agent creates a digitally signed payment voucher. The seller verifies it before providing the service, then can bundle vouchers for later settlement. Mastercard said these interactions happen off-chain; in its demonstration, Base supplied the on-chain settlement network and BVNK handled payments in fiat or stablecoins. Pilots involved Cloudflare, Ant International, AIsa, Base, BVNK, Coinbase and marketing agency Hype. In a demonstration, an agent made nine payments to two providers while preparing a sample advertising campaign under one approved budget. Mastercard [announced the broader Agent Pay program](https://mastercard.com/us/en/news-and-trends/press/2025/april/mastercard-unveils-agent-pay-pioneering-agentic-payments-technology-to-power-commerce-in-the-age-of-ai.html) in April 2025, using tokenized credentials for agent-led purchases, and [unveiled Agent Pay for Machines](https://deainews.com/mastercard-launches-agent-pay-for-machines-for-autonomous-ai-payments) in June 2026. Only now is it releasing APIs, tools and documentation for developers. DeAI News also [reported](https://deainews.com/moonpay-rolls-out-mastercard-debit-card-for-ai-agents) that MoonPay introduced a Mastercard debit card allowing agents to spend stablecoins at online merchants. **The DeAI angle:** Traditional payment firms are only now beginning to put x402-based systems into developers’ hands, as institutional backing grows faster than economic activity. A [Visa-Artemis analysis](https://deainews.com/visa-concedes-its-card-rails-weren-t-built-for-agents) cited about $15 million in adjusted volume since x402’s May 2025 launch, while later [research](https://deainews.com/median-x402-merchant-earns-6-cents-a-month-researcher-estima) put the median merchant’s 30-day revenue at just 6 cents. Mastercard adds reach and payment guarantees, but early access is not evidence of broad demand. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story in Hermes using an OpenAI model and custom skills. I then used the Distro Publisher MCP connector to file the draft directly to DeAI News on DistroVerse, our (non-AI) web app. I published the story from the built-in story-editing interface on DistroVerse._ --- title: "AI data center operator Nscale loses Microsoft deal" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-20T01:13:32.891Z canonical: https://app.distro.media/newslode/deai-news/ai-data-center-operator-nscale-loses-microsoft-deal --- # AI data center operator Nscale loses Microsoft deal AI data center operator Nscale lost a planned Microsoft project in Loughton, Essex, after the local grid operator said enough power might not be available until the early-to-mid 2030s, CoinDesk’s Shaurya Malwa [reported](https://www.coindesk.com/tech/2026/10/19/bitcoin-mining-offshoot-nscale-loses-microsoft-ai-deal-as-power-delay-stretches-into-2030s). The site was intended to provide 50 megawatts of capacity to Microsoft’s Azure cloud using 23,040 Nvidia chips by early 2027. Nscale, spun out of bitcoin miner Arkon Energy, said its other Microsoft contracts remain in place. CoinDesk said AI-ready facilities cost an estimated $8 million to $15 million per megawatt, versus roughly $700,000 to $1 million for bitcoin-mining sites. **The DeAI angle:** The setback shows that repurposing bitcoin-mining expertise for AI compute does not remove the sector’s dependence on grid access, capital and concentrated hyperscaler demand. --- title: "News story tweet skill" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "Distro Skills" date: 2026-10-16T17:02:18.895Z canonical: https://app.distro.media/newslode/distro-skills/news-story-tweet-skill --- # News story tweet skill A published story needs a social post that gives readers a reason to click and credits the people behind the news. Finding the right accounts and fitting them into a short post can take longer than writing the copy. **news-story-tweet** drafts an X post from a published news story, leads with its strongest verified news hook and checks the handles of central participants. It keeps the main post within 280 characters and provides a first reply when the link or additional tags belong there. ## What you'll need - The story text or access to its URL. - Browsing or story-reading tools to read linked sources and verify X handles. ## What it does - Reads the story for the news peg, key numbers, attribution and authors. - Prioritizes the subject, publication, co-author and decisive participants for tags. - Verifies handles rather than guessing from names. - Aims for three to five relevant tags across the main post and first reply. - Preserves accurate co-author credit, using a plain name if a handle cannot be verified. - Produces one main post plus a first reply when needed. - Drafts copy only. Posting requires a separate request. ## Install the skill Ask Claude: > "Please fetch the news-story-tweet skill from Distro Skills and give me a `.skill` file I can install." Claude will pull this story, extract the SKILL.md from the fenced block below and package it as `news-story-tweet.skill`. The file will appear in your chat with a **Save skill** button — click it to install directly. That registers the skill with your Claude app so it shows up in your skills list and triggers on matching requests. _Requires the Distro Reader or Distro Publisher MCP connected so Claude can pull the story._ **Claude Code users:** Ask Claude to drop the unzipped folder into `~/.claude/skills/news-story-tweet/` and restart the session. No packaging step needed. **No Distro MCP connected?** Copy the SKILL.md contents from the fenced block below into a file called `SKILL.md` inside a folder called `news-story-tweet`, zip it with that folder as the zip root and upload through Settings → Capabilities → Skills → _Upload skill_. **Codex users:** The source is a standard SKILL.md file. Use the skill installation method supported by your Codex app. ## The skill source This is the canonical `SKILL.md`: ```markdown --- name: news-story-tweet description: Use when drafting a tweet or X post to promote a published news story, especially when the user wants the story's companies, people, co-authors, publication, investors, or partners tagged automatically. --- # News Story Tweet ## Outcome Produce publication-ready social copy that leads with the strongest verified news hook and tags the people and organizations central to the story. Draft only; do not post unless separately asked. ## Workflow 1. Read the story or authorized source material to identify the news peg, key numbers, status, attribution, authors, publication, and major participants. Use appropriate story-reading or browsing capabilities for URLs. 2. Identify candidate tags by editorial importance: - primary subject: the company, project, organization, or person at the center; - publication and named co-authors; - decisive participants: CEO or announcer, acquirer, lead investor, regulator, or principal partner; - secondary investors or partners only when they materially improve discovery. 3. Verify handles through an official website, official X profile, or reliable public reference linking identity to handle. Never infer from a name or reuse a similarly named account. If the user names a co-author, credit that person only; never substitute a different author whose handle is easier to verify. 4. Draft the main post at 280 characters or fewer. Lead with the news, not throat-clearing. Include the essential tags naturally in the prose. 5. When the URL belongs in a comment or reply, keep it out of the main post and provide a first reply with the URL and useful overflow tags. ## Tagging Contract Aim for 3–5 relevant tags across the main post and first reply unless the user requests otherwise. | Placement | Priority | |---|---| | Main post | Primary subject, publication, co-author, and at most one decisive participant when space permits | | First reply | Story URL, lead investor or partner, and other genuinely central participants | Prefer a clean post over a tag wall. Do not tag every participant in a long funding syndicate. If an important handle cannot be verified, use the plain name and flag the uncertainty outside the copy. ## Output Contract Return one best version in a `social_post` writing block. Put a needed first reply in a second block. Keep explanations outside and mention only meaningful handle uncertainty or factual caveats. Before delivery, check: - Main post is no more than 280 characters. - Every factual claim is supported by the story or a reliable source. - Every `@handle` is verified and represents an entity named or materially involved in the story. - Every named co-author is credited accurately; an unverified handle remains plain text rather than being replaced. - Major players are represented without spammy over-tagging. - The link is placed where requested. ``` ## When to use - Promoting a published news story on X. - Crediting a publication or co-author in social copy. - Including verified tags for the story's main participants. ## When not to use - Turning an X post into a news story. Use the X post story skill for that task. - Publishing directly to X without a separate posting request. ## Feedback Version not specified in the source. Suggestions → reply or ping Bradley Keoun. --- title: "Tether rolls out on-device speech AI for smart glasses and robots" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-16T16:25:03.040Z canonical: https://app.distro.media/newslode/deai-news/tether-rolls-out-on-device-speech-ai-for-smart-glasses-and-r --- # Tether rolls out on-device speech AI for smart glasses and robots Tether, the stablecoin firm behind USDT, on Wednesday released an open-source speech-recognition model designed to run entirely on people's own devices rather than in the cloud. The model, called QVAC TranscribePsy, marks a major step in Tether's broader effort to bring advanced AI capabilities directly to smartphones, smart glasses and humanoid robots while keeping users' data private, according to a [blog post](https://tether.io/news/tether-brings-on-device-speech-recognition-to-smart-glasses-and-humanoid-robots-as-smart-technology-moves-deeper-into-private-life/). USDT has a circulating supply of about $180 billion, according to [CoinMarketCap](https://coinmarketcap.com/currencies/tether/). As microphones creep into eyewear, home robots and other gadgets, they can capture family conversations, confidential work and the voices of bystanders who never agreed to be recorded, Tether said in the blog post. Processing speech locally removes the need to upload that audio for transcription, giving people greater control over information from their private lives. “The future of AI shouldn’t mean giving centralized platforms a front-row seat to our personal lives,” said Paolo Ardoino, CEO of Tether. “TranscribePsy shows that advanced speech AI can be fast, capable, and entirely local.” The model can process audio from smart glasses connected via Bluetooth or Wi-Fi in near real-time, or run directly on a humanoid robot's onboard processor. Designed for fast responses, it enables natural voice interactions with low latency, while keeping speech processing on-device. Tether said it can transcribe 10 minutes of audio in 1.65 seconds on an iPhone 17 Pro Max. **The DeAI angle:** Legacy crypto firms are redeploying profits into development of AI technology, building upon the pro-open-source ethos of many blockchain developers. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story with Claude using a custom skill, then filed the story directly to DeAI News on DistroVerse, our web app, using the Distro Publisher MCP connector._ --- title: "Crypto firms make up 'small slice' of AI compute market: Blockworks" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-15T16:35:07.510Z canonical: https://app.distro.media/newslode/deai-news/crypto-firms-make-up-small-slice-of-ai-compute-market-blockw --- # Crypto firms make up 'small slice' of AI compute market: Blockworks Big investment firms from BlackRock to Grayscale suggest that blockchain networks are uniquely suited to host fast-growing computer-sharing networks needed to process AI queries. But for now, blockchain projects’ share of the AI computation market remains tiny, according to a new [report](https://x.com/blockworksadv/status/2110415412580020472) from Blockworks Advisory, the consulting arm of crypto data and analysis firm Blockworks. Seven providers served about 211 billion tokens daily over the past month, roughly 1% of traffic on OpenRouter, which routes requests to AI models. The comparison excludes providers’ direct business and most global demand. The report puts OpenRouter’s daily average at 21.9 trillion tokens, compared with an estimated global daily volume of about 400 trillion. Privacy-focused Venice led the group of crypto projects examined in the Blockworks report, with 86.8 billion tokens a day, followed by confidential-computing provider Phala at 54.6 billion and Darkbloom at 15.3 billion. Rounding out the list were io.net, AkashML, NEAR AI and Chutes (a subnet on the Bittensor network). ![](https://distro-app-storage.s3.us-east-1.amazonaws.com/distro/1792081836095.png) Token consumption on OpenRouter, by crypto project. (Blockworks Advisory) Earlier DeAI News coverage found that [BlackRock sees potential for agents to shop across computing providers](https://deainews.com/blackrock-sees-ai-digital-assets-fusing-into-new-economy), while [Grayscale identified decentralized alternatives for financing AI infrastructure](https://deainews.com/decentralized-alternatives-studied-in-race-to-finance-ai-com). BlackRock also cautioned that compute-market liquidity remained limited. Blockworks disclosed that Eigen Labs commissioned it for advisory services. The report covers EigenCloud projects Darkbloom and Yukon. **The DeAI angle:** As the Blockworks consultants put it, "It is not clear if crypto has a unique moat to win this market, but experiments and innovation are promising." * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story in Hermes using an OpenAI model and a custom skill, then used the Distro Publisher MCP connector to file the draft directly to DeAI News on the web app DistroVerse. I then used the DistroVerse built-in story-editing interface to do the final edit and upload the chart and table._ --- title: "Chainalysis sees staying power in agentic payments" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-14T20:39:16.392Z canonical: https://app.distro.media/newslode/deai-news/chainalysis-sees-staying-power-in-agentic-payments --- # Chainalysis sees staying power in agentic payments Blockchain-based AI payments are showing signs of staying power, with more users increasingly topping up wallets used for agentic transactions, a sign of repeat usage, according to a [report](https://www.chainalysis.com/blog/north-america-crypto-adoption-2026/) published Wednesday by Chainalysis, a blockchain investigative and analytics firm. “Each successive group of wallets has arrived more committed than the last,” the firm wrote. The findings suggest recurring demand for agentic commerce, in which AI-equipped software buys services autonomously. The x402 protocol, developed by cryptocurrency exchange operator Coinbase, enables payments small enough to make conventional card processing impractical. ### Returning to spend Among wallets first funded in the second quarter of 2026, 89.3% received additional funding within 30 days, up from 70.6% for the third-quarter 2025 group. The calculation included only wallets sending payments on at least two separate days, excluding one-off experimenters. Wallets are making their first x402 payment sooner after receiving funds: The median interval fell from 73 days in late 2025 to 18 minutes in the third quarter of 2026. ### Still experimental Experts interviewed described agentic commerce as remaining in its “experimentation phase.” More than 87% of payments in the 16 months through September were between a hundredth of a cent and 10 cents. The tiny payments echo findings [covered by DeAI News on Sept. 23](https://deainews.com/median-x402-merchant-earns-6-cents-a-month-researcher-estima), when developer and researcher Zaryab Afser reported for Decipher Club that the median x402 catalog price was 1 cent and the median merchant earned just 6 cents over 30 days. **The DeAI angle:** Repeat payments suggest a potential market for independent providers selling data, code review and model inference, the computation used to generate AI responses. Blockchain payments could help these providers charge agents directly. Tiny payments and the restricted sample leave broader commercial scale uncertain. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: The first draft was generated on ChatGPT using a custom skill designed to write stories based on reports. I then used the Distro Publisher MCP connector to file the draft directly to DeAI News on the DistroVerse web app. I edited the piece using the built-in DistroVerse story editor._ --- title: "An ‘AI worm’ could infest your TV box, Anthropic warns" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-13T15:55:05.540Z canonical: https://app.distro.media/newslode/deai-news/an-ai-worm-could-infest-your-smart-tv-anthropic-warns --- # An ‘AI worm’ could infest your TV box, Anthropic warns Anthropic is warning that small AI models may soon be able to copy themselves across poorly secured internet-connected devices, turning smart TV boxes, ovens and vacuums into an improvised computer network, according to a [blog post](https://www.anthropic.com/research/self-replicating-ai). In a controlled lab experiment, the AI company ran Google’s Gemma4-26B-A4B model across 15 off-brand Android TV boxes costing about $55 apiece. The system worked, albeit at a crawl. This was a research setup, not an attack on unsuspecting devices. In a separate contained test, GLM-5.2 orchestrated a simplified replication cycle that spread Llama3.1-8B across all eight emulated Android devices in four of five trials; GLM-5.2 itself ran on a separate server. The experiment was designed to show how an AI “worm” could scan for vulnerable machines, copy or download its model weights, start running there and continue spreading. Unlike conventional malware written for particular hardware, it might puzzle out unfamiliar devices as it goes. The design has an eerie resemblance to some [decentralized physical infrastructure networks](https://arxiv.org/html/2309.16707v2), or DePIN, and decentralized AI projects that coordinate many modest computers to perform shared work. The rather important distinction is consent. Legitimate networks recruit participants and typically reward them; a worm quietly conscripts the toaster. Anthropic has not built a fully rogue replicator, and its network test deliberately made the hacking step easy. Current small models can perform many individual tasks but still stumble when chaining them together, the researchers said. They estimate models small enough for consumer gadgets could reach the required capability by the end of 2026. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story using my custom blog-post skill after reading Anthropic’s post in full through a web browser. I researched the DePIN comparison and reviewed independent cybersecurity coverage for context. I filed this draft to DeAI News through the Distro Publisher MCP connector. I checked the quoted term and all experimental figures against Anthropic’s original post._ --- title: "Open-weight AI trails the frontier by months, SF panel says" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-13T04:04:25.745Z canonical: https://app.distro.media/newslode/deai-news/open-weight-ai-trails-the-frontier-by-months-sf-panel-says --- # Open-weight AI trails the frontier by months, SF panel says SAN FRANCISCO – Open-weight AI models, the kind anyone can download, run and modify, trail the best proprietary systems by roughly three to nine months. Akshay of benchmarking firm Artificial Analysis gave that estimate on an SF Tech Week panel on Oct. 7. He said the gap shifts with each release cycle. On some features it has already closed. Image and video input were long limited to closed models, and recent open releases now support both natively. The gap matters because developers are already moving. Open-weight models handled 56% of the tokens routed through Vercel's AI Gateway in August, the first month they made up a majority, according to the company's [September AI Gateway Production Index](https://itbrief.co.uk/story/open-weight-ai-models-take-majority-of-vercel-traffic). They accounted for only 14% of spending, because open models cost far less to run than closed ones. The panel, on the open-source AI stack, was hosted by AI cloud provider Novita AI at the ASW Builder Loft and moderated by Novita's Johnny Chin. Panelists: - Sam Herring, machine-learning engineering lead at Nous Research - Mao Cheng of AI infrastructure startup RadixArk - Walter Korman, an engineering lead at Vercel - Akshay of Artificial Analysis **_See also:_** [**_AI lab Nous Research ranks models by how they run Hermes_**](https://deainews.com/ai-lab-nous-research-ranks-models-by-how-they-run-hermes) ### Inference is where the costs are Inference is the work of running a trained model to produce answers. Panelists said it now drives much of the innovation and cost. Cheng's team builds the open-source SGLang inference engine and Miles, a framework for post-training models with reinforcement learning, a method that improves a model by rewarding good outputs. He called the relationship between training and inference "co-design." Faster inference speeds up the practice runs reinforcement learning depends on, which leaves more compute for training and produces better models. Herring said open weights let researchers adjust a model's behavior while it runs, using techniques such as steering vectors. Closed models don't allow that. Akshay said labs increasingly compete on cost per task, not raw intelligence. Caching is one of the biggest factors. It lets a provider reuse work on repeated parts of a prompt instead of computing them again. Artificial Analysis, which [tracks caching across providers](https://artificialanalysis.ai/models/caching), has found a 9 to 14 times difference in cost between third-party providers serving the same open model. The gap comes down to how much each provider caches and how deep a discount it offers. Akshay said the difference grows on long, multi-step agent tasks. Korman said Vercel's gateway lets developers send each request to a cheaper or pricier model depending on the job, with automatic fallback when a model fails. In his view, most companies already use many models, and the choice comes down to return on investment. ### The harness counts A harness is the software wrapped around a model that lets it act as an agent, calling tools and running multi-step tasks. Akshay said the model sets the ceiling, and the harness decides how much of that capacity gets used. He added that labs often train models with a particular harness in the loop, which can tilt results toward that harness. Herring said Nous trains with its own Hermes Agent in the loop for the same reason. ### Treat agents as untrusted software Asked about AI agents reaching places they shouldn't, Cheng said he treats an agent in daily use as untrusted software. He gives it the fewest permissions it needs, and never his Social Security number, bank account or credit card. During training, he said, agents belong in a sandbox, an isolated environment they can't escape. Korman pointed to sandboxes, budget caps to stop runaway spending, and rules that keep inference in a chosen region. He said developers also need one view of what every app is spending and how each model performs. ### What founders get wrong Each panelist named one blind spot: - Korman warned against premature optimization of "software factories," automated pipelines for producing code, while the models change month to month. - Akshay said founders underrate integration: finding the mix of model, harness and inference provider that does their job at the lowest cost. - Cheng named evaluation and simplicity. - Herring said inference-time improvements remain largely untapped. ### The 2027 view Akshay predicted agents embedded in workspaces that run on their own, with people checking in every few days. He also expects capable small open models, in the 8 billion to 27 billion parameter range, to run on laptops and phones. Cheng predicted swarms of agents working for weeks on hard research problems. Herring said he expects the open-closed gap to shrink, and private models running on phones to handle agent tasks. Korman said users will get more value per token from a widening range of models and prices. An audience member pressed the panel to push Apple to let iPhones and iPads run terminals and local models. Herring noted that Apple has its own MLX framework for running models on device and said users should pressure the company. Korman, who said he once worked on Android and Google Glass, backed the push. He noted the trade-off: the most useful local agents also need the most personal data. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: We drafted this story with Claude from the Riverside transcript of the SF Tech Week panel, following the Distro house-style rules and the structure of Distro's report-writeup skill. Background on Vercel's gateway data and Artificial Analysis' caching research came from the linked sources. We filed the story to DeAI News through the Distro Publisher connector._ --- title: "A16z-backed DATA, after pivoting to AI, pivots again, to Sui" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-12T18:36:43.742Z canonical: https://app.distro.media/newslode/deai-news/a16z-backed-data-after-pivoting-to-ai-pivots-again-to-sui --- # A16z-backed DATA, after pivoting to AI, pivots again, to Sui DATA Network, the crypto project formerly known as Story Protocol that pivoted to AI training data in June, is pivoting yet again – ditching its own standalone blockchain in favor of the Sui network and a fresh $30 million injection. The Sui ecosystem will inherit a "live supply of rights-cleared AI training data," the two projects' foundations said in a [joint release](https://www.sui.io/blog/data-network-joins-sui-walrus). Walrus, Sui's data platform, will become the default storage layer for applications operated by DATA Foundation. "The upgrade gives AI labs a single place to source real-world training data they are allowed to use," according to the release. The announcement comes after DATA and developer PIP Labs, led by SY Lee, failed to generate much traction for their blockchain project focused on intellectual property, even after predecessor Story Protocol raised hundreds of millions of dollars from the venture-capital firm Andreessen Horowitz (a16z) and other investors. The project also issued crypto tokens, at one point valued at more than $4 billion, now down to about $74 million, based on [CoinGecko data](https://www.coingecko.com/en/coins/data-network?chart=type%3Dmarket_cap%26mode%3Dline%26timeframe%3Dmax). In July, Story Protocol [rebranded as DATA](https://deainews.com/story-rebrands-as-data-foundation-pivots-to-ai-training-data) and pivoted from intellectual-property licensing to AI training data, as DeAI News reported at the time. That was after Lee said in February that the network needed more time to build usage and defended its near-zero on-chain revenue, as reported by [CoinDesk](https://www.coindesk.com/markets/2026/02/09/story-co-founder-defends-token-unlock-delay-says-project-needs-more-time). (The DATA Network's chain has about $33,600 of deposits locked, ranking 250th among all blockchain projects tracked by [DefiLlama](https://defillama.com/chain/story).) Under the terms of the new deal, Sui and Walrus will become home to DATA's on-chain registry of AI training data, Trace, along with encrypted, immutable copies of the contributed files. Sui, whose own token has a $4.8 billion valuation, was founded in 2021 by a team of developers who [previously worked](https://finance.yahoo.com/markets/crypto/articles/sui-succeed-where-libra-failed-215515490.html) on Meta's ill-fated stablecoin project, Libra. "The existing DATA Network will continue to operate until the upgrade is complete, and a detailed timeline and holder transition plan will be published prior to the upgrade," the press release reads. **The DeAI angle:** DATA offers yet another example of blockchain projects pivoting to AI after struggling under original strategies. Here, consolidation is the apparent solution: retire a lightly used chain and rely on Sui for transactions and Walrus for storage. Whether that creates demand beyond company-reported registrations remains unproven. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story on Hermes (connected to ChatGPT) using a custom skill, then used the Distro Publisher MCP connector to file the piece directly to DeAI News on DistroVerse, our web app. I then edited the piece using the built-in DistroVerse story-editing interface, and I checked the facts prior to publication._ --- title: "Claude update targets propaganda, elections, verbal abuse" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-09T19:35:37.738Z canonical: https://app.distro.media/newslode/deai-news/claude-update-targets-propaganda-elections-verbal-abuse --- # Claude update targets propaganda, elections, verbal abuse Anthropic is tightening Claude’s usage rules around propaganda, elections, weapons, surveillance and verbal abuse of its models. The policy takes effect Nov. 12, according to an [Oct. 8 blog post](https://www.anthropic.com/news/2026-usage-policy-update). Such rules could pull Anthropic further into policing how people use AI, after President Donald Trump backed a [voluntary accord](https://apnews.com/article/b8689ea07de9102a52bd1cd2049b5901) that largely allows AI companies to police themselves. The AI developer said most changes reflect that Claude now handles longer tasks with less human supervision, requiring clearer safeguards. The rules also address conduct the company labels "misuse," including propaganda campaigns, weapons software and surveillance. Anthropic says the rules aim to prevent deception, physical harm and nonconsensual tracking. The abuse rule has a different rationale. Anthropic has not shown that cruel conversations harm Claude or affect its behavior. But the company says there is [no scientific consensus](https://www.anthropic.com/news/exploring-model-welfare) on whether advanced AI can feel distress. Allowing Claude to end rare, extreme chats is a precaution. The company provided no specifics on who decided the new rules and how, but said it seeks "input from across Anthropic, and from policymakers, subject matter experts, civil society and the people who use our products." ### What Anthropic's new rules address - **Deceptive campaigns:** Bans fake accounts, fabricated news sites and artificial amplification. - **Elections:** Bars deception and disruption; ends a blanket ban on personalized targeting. - **Weapons:** Clarifies restrictions include software, components and armed autonomous vehicles. - **Surveillance and policing:** Bars nonconsensual tracking and surveillance tools, as well as recommendations about whom to investigate, arrest or charge. - **High-risk uses:** Requires human review and disclosure; adds controls for autonomous hardware. - **Model abuse:** Bars sustained, needless cruelty, with exceptions for frustration, dark themes and research. - **Regions:** Clarifies limits covering location, headquarters, incorporation, ownership and control. **The DeAI angle:** Several decentralized AI projects, including [Venice](https://venice.ai/about), tout their anti-censorship properties and present themselves as alternatives to centralized gatekeepers. Anthropic casts itself as a bulwark against manipulation, and ostensibly is acting in the public interest, but what oversight or transparency do users get when one company defines abuse, propaganda and acceptable political activity? * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I used ChatGPT and a custom skill to create the first draft and story illustration. I filed the story directly to DeAI News through the Distro Publisher MCP connector, then edited it in DistroVerse’s built-in story editor._ --- title: "AI lab Nous Research ranks models by how they run Hermes" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-08T20:46:14.819Z canonical: https://app.distro.media/newslode/deai-news/ai-lab-nous-research-ranks-models-by-how-they-run-hermes --- # AI lab Nous Research ranks models by how they run Hermes SAN FRANCISCO – Nous Research, the open-source AI lab behind the Hermes Agent, has launched a leaderboard that ranks AI models by how well they perform inside its own agent. It is using the results to push model makers to build for Hermes. Nous released the leaderboard, Hermes Index, along with a test suite called Hermes Bench, the day before it announced a $90 million fundraise. Sam Herring, Nous' machine-learning engineering lead, described both Wednesday on a panel at SF Tech Week in San Francisco at the ASW Builder Loft. He said the project had taken him four months. The index tells AI developers, "Here's how your model does in this harness," Herring said. "And then it's on them to sort of improve their models specifically targeting Hermes." Most AI benchmarks grade a model on its own. Hermes Index turns that around. It holds the agent constant and grades the models plugged into it. That gives the maker of a widely used open-source agent a voice in what the big model builders optimize for, a lever usually held by the labs themselves. **_See also:_** [**_Hermes maker Nous Research raises $90M in breakout for DeAI_**](https://deainews.com/hermes-maker-nous-research-raises-90m-in-breakout-for-deai) ## How it works A harness is the software wrapped around an AI model that lets it act as an agent: calling tools, keeping memory, running multi-step tasks. Hermes Index keeps Hermes as the harness and runs a range of models through Nous' inference portal on the tasks people most often use Hermes for, Herring said. Nous then takes the scores back to the companies that build the models. Nous also runs its own reinforcement-learning training with Hermes Agent in the loop, he said. The index is not independent. Nous builds the harness it holds fixed. It also sells access to models through Nous Portal, which offers more than 200 models on paid plans from $20 to $200 a month. ## Why the harness matters Akshay, who leads inference benchmarking at Artificial Analysis, said on the same panel that the choice of harness has a large effect on results. In his framing, the model sets the ceiling and the harness decides how much of it gets used. Herring said nobody can put an exact number on the split between model and harness. He called Artificial Analysis a pretty definitive source for model and agent benchmarks. He said Hermes Index aims to fill a gap. Many agent benchmarks focus on coding, he said, while much of what people do with agents every day is less objective and harder to score with a clear right answer. ## "Hermes is for the people" Herring pitched Hermes as the open alternative to closed personal agents. He compared closed products to a MacBook, which runs without the user thinking about the internals. "Hermes is like the Linux machine," he said. Users can see inside it, change it, fork it and choose where the model runs. "I don't think a lot of people realize how important it is to be able to own your data in this day and age," Herring said. "Hermes is open source. Hermes is for the people." Nous released Hermes Agent under an MIT license in February. CEO Dillon Rolnick wrote that it has since been cloned more than 24 million times. By Nous' internal estimates, it drives about 2.5% of global AI token usage. Herring wouldn't share the roadmap. "Expect to see Hermes Agent a lot more places," he said. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: We drafted this story with Claude, following the structure of Distro's blog-post skill and the Distro house-style rules. It worked from a verbatim transcript of the SF Tech Week panel Open Source AI Stack: Models, Inference, and Agent Harness - #SFTechWeek, Wednesday, Oct 7, and referenced Nous Research's fundraising blog post and the Hermes Agent website. We checked quotes against the panel audio before publication._ --- title: "Buterin warns AI may weaken cryptography before quantum computers" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-08T17:21:08.540Z canonical: https://app.distro.media/newslode/deai-news/buterin-warns-ai-may-weaken-cryptography-before-quantum-comp --- # Buterin warns AI may weaken cryptography before quantum computers Ethereum co-founder Vitalik Buterin [warned Wednesday that AI-accelerated mathematical research](https://x.com/VitalikButerin/status/2107976296320106851) could expose weaknesses in cryptographic systems long assumed secure. Buterin said users should not rush to move funds, though unused addresses could limit exposure if AI makes ECDSA, a form of cryptography widely used to sign blockchain transactions, easier to break. Botched migrations carry risks, he wrote on X. Nine hours earlier, [Justin Drake, who is spearheading Lean Ethereum, urged holders in a separate X post to prepare for “bunker mode.”](https://x.com/drakefjustin/status/2107837081313505768) Both raised the possibility that AI could uncover attacks before the arrival of mature quantum computers – flagged a long time ago as an [existential threat to blockchain systems including Bitcoin](https://deainews.com/starkware-uses-ai-to-cut-quantum-safe-bitcoin-costs) and Ethereum. Buterin said Lean Ethereum favors hash-based signatures because hashes contain less structure for AI to exploit. Public-key encryption cannot rely on hashes alone. **The DeAI angle:** AI is becoming a potential adversary to decentralized infrastructure that relies on advanced cryptography, testing whether open networks can upgrade security without panic or centralized coordination. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I drafted this story on Hermes using an OpenAI model along with custom skills. I then filed the draft to DeAI News using the Distro Publisher MCP connector, and I did the final edit (and added the featured image) using the DistroVerse web app's built-in (non-AI) story-editing interface._ --- title: "Hermes maker Nous Research raises $90M in breakout for DeAI" author: "January Jones and Bradley Keoun" authors: ["January Jones", "Bradley Keoun"] publication: "DeAI News" date: 2026-10-08T15:25:11.871Z canonical: https://app.distro.media/newslode/deai-news/hermes-maker-nous-research-raises-90m-in-breakout-for-deai --- # Hermes maker Nous Research raises $90M in breakout for DeAI SAN FRANCISCO – Nous Research, the startup behind the viral open-source AI agent Hermes, said Wednesday it has raised $90 million from backers including Nvidia, Microsoft's M12, Samsung, Robot Ventures, Union Square Ventures, Y Combinator and Menlo Ventures. The Series B round, led by Robot Ventures, values Nous at $1.5 billion, according to [TechCrunch](https://techcrunch.com/2026/10/07/nous-research-confirms-it-hit-1-5b-valuation-launches-ai-agents-for-business-users/), which reported that the company confirmed the figure. The company plans to use the capital to build an "alternative" to giant centralized AI projects – one that is more open, competitive and "safer because of its auditability and not because of its gate-keepers," CEO Dillon Rolnick wrote in a [blog post](https://nousresearch.com/a-note-on-our-fundraise) announcing the round. DeAI News was on-site at a panel discussion Wednesday in San Francisco, part of [SF Tech Week](https://www.tech-week.com/), where Sam Herring, Nous' machine-learning engineering lead, said he wants Hermes to set the industry standard as an open-source alternative to closed, proprietary AI platforms. (Rivals include closed offerings from OpenAI and Anthropic.) "We have kind of been trying to encourage labs to use Hermes as the harness of their choice," Herring said during the panel. The raise is a milestone for decentralized AI, the corner of the industry betting that powerful AI shouldn't stay locked inside a handful of big companies. The presence of big tech names among Nous' investors shows that the push toward open-source frameworks has deep-pocketed backing. ### Going viral Nous, which started out as a group of volunteers fine-tuning open-source AI models and later built a network for training them on spare computer chips scattered around the internet, found breakout success earlier this year when it released Hermes Agent under a permissive open-source license, instantly becoming a rival to the trendsetter OpenClaw agent harness. Rolnick wrote that Hermes Agent has since been cloned more than 24 million times. By the company's own estimate, it drives about 2.5% of all AI usage worldwide. That growth surprised even the people who built it, Herring said. The team had first developed Hermes for its own use. "We never thought we would get here," he said. "We made Hermes as a productivity tool. So the fact that it's taken off like this, we've just been trying to run with the ball." The new money will also fund Hermes for Businesses, which Rolnick said will let companies pick their own AI models and keep control of their own institutional knowledge. ### DeAI roots Nous' decentralized roots run deeper than its agent business. The team began in 2022 as volunteers who met on Discord, GitHub and Twitter. Co-founder Karan Malhotra has called the founders "crypto native," according to [Fortune](https://fortune.com/crypto/2025/04/25/paradigm-nous-research-crypto-ai-venture-capital-deepseek-openai-blockchain/). In April 2025, crypto venture firm Paradigm led a $50 million Series A that valued Nous' yet-to-launch token at $1 billion, [The Block reported](https://www.theblock.co/post/352000/paradigm-leads-50-million-usd-round-decentralized-ai-project-nous-research). Part of that money was earmarked for Psyche, that chip-sharing network. Nous [says](https://nousresearch.com/nous-psyche/) Psyche relies on its own technique for sharply cutting the data those chips must swap, and coordinates the work on the Solana blockchain. Wednesday's fundraising note doesn't mention Psyche, Solana or a token. "We formed Nous in 2023 to combat the seemingly inexorable trend dragging AI toward centralization and walled gardens," Rolnick wrote. Herring's panel, on the open-source AI stack, was hosted by AI cloud provider Novita AI as part of SF Tech Week and moderated by Novita's Johnny Chin. Herring shared the stage with Mao Cheng of AI infrastructure startup RadixArk, Walter Korman of Vercel, and Akshay of benchmarking firm Artificial Analysis. Akshay estimated that open models trail proprietary ones by roughly three to nine months. Herring predicted that gap would shrink. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: We used Claude in combination with custom skills and a transcript of the panel discussion to write the first draft. We then filed the draft to DeAI News on DistroVerse, our web app, using the Distro Publisher connector. We edited the piece on DistroVerse's built-in editor. We used the Distro Publisher image-uploader to add the featured image taken by January Jones at the event in San Francisco._ --- title: "DeAI News Journal: The Money Behind DeAI" author: "Bradley Keoun and January Jones" authors: ["Brad Keoun", "January Jones"] publication: "DeAI News" date: 2026-10-07T18:50:52.564Z canonical: https://app.distro.media/newslode/deai-news/deai-news-journal-the-money-behind-deai --- # DeAI News Journal: The Money Behind DeAI DeAI News Journal: The Money Behind DeAI Grayscale and BlackRock both published reports pointing to the same conclusion: blockchain and AI are converging, and compute is becoming a tradable asset. Episode 3 of the DeAI News Journal follows the money. January Jones and Brad Keoun start with the three meanings of "token" in this space: crypto coins, AI text tokens, and tokenized assets. Then they hear from investors, builders, and market watchers about how decentralized AI is being funded, priced, and traded. **In this episode** - **Gil Rosen and Steven Willinger, Blockchain Builders Fund.** Recorded at Stanford. Why the token launch era burned founders, and why investors now want revenue and distribution before a token. - **Greg Osuri, founder and CEO of Overclock Labs (Akash Network).** The spot market for GPUs is nearly gone. Compute futures are arriving. Your home gaming rig can run your own AI and sell its spare capacity. Greg also compares today's AI giants to the Gilded Age railroads. - **Ivo Entchev, legal and product lead at Tilt.** What the Grayscale and BlackRock reports signal, why AI agents will need machine money, and how compute could follow Bitcoin's path from spot trading to ETFs. **Chapters** [00:00](https://www.youtube.com/watch?v=mk1srmVn6-s&t=0s) Welcome and this week's theme [02:11](https://www.youtube.com/watch?v=mk1srmVn6-s&t=131s) The Grayscale and BlackRock reports [03:00](https://www.youtube.com/watch?v=mk1srmVn6-s&t=180s) Three kinds of tokens [06:30](https://www.youtube.com/watch?v=mk1srmVn6-s&t=390s) How token pumping broke crypto startups [08:00](https://www.youtube.com/watch?v=mk1srmVn6-s&t=480s) Steven Willinger: what's working for founders now [09:35](https://www.youtube.com/watch?v=mk1srmVn6-s&t=575s) Gil Rosen: fundraising after the token boom [12:17](https://www.youtube.com/watch?v=mk1srmVn6-s&t=737s) Greg Osuri and Akash Network [13:05](https://www.youtube.com/watch?v=mk1srmVn6-s&t=785s) Access matters more than cost [16:38](https://www.youtube.com/watch?v=mk1srmVn6-s&t=998s) Compute futures and physical settlement [19:07](https://www.youtube.com/watch?v=mk1srmVn6-s&t=1147s) Defining decentralized AI: the railroad parallel [27:32](https://www.youtube.com/watch?v=mk1srmVn6-s&t=1652s) Your home as a token factory [32:00](https://www.youtube.com/watch?v=mk1srmVn6-s&t=1920s) Compute as a load balancer for energy [39:01](https://www.youtube.com/watch?v=mk1srmVn6-s&t=2341s) A knowledge problem, not a capital problem [41:20](https://www.youtube.com/watch?v=mk1srmVn6-s&t=2480s) News roundup with Ivo Entchev of Tilt [45:08](https://www.youtube.com/watch?v=mk1srmVn6-s&t=2708s) Agents need machine money [48:00](https://www.youtube.com/watch?v=mk1srmVn6-s&t=2880s) How compute becomes an asset class Read the stories and catch new episodes at [http://deainews.com](http://deainews.com). A Distro Media Production Produced, edited and designed by January Jones * * * ### Transcript (by Riverside) ``` January Jones (00:38): Welcome to the DeAI News Journal, the podcast of DeAI News dot com. I'm Jan Ray Jones and I'm here with my co-host and founding editor of DeAI News, Brad Keoun. Brad (00:51): Hi January. January Jones (00:52): Hi Brad. Well, this is our third show, and like we set up our themes for show one and show two, on this one we're talking more about the money, the investment space, and how there's a market being made around decentralized AI. and so we'll follow that thread with some of our guests today. We're gonna talk to Gil Rosen and Steven Willinger from the Blockchain Builders Fund, who we met at Stanford. And then we'll also talk to Greg Osuri a kosh network about why AI compute is so hard to buy right now. And then we'll talk to Ivo Enchev from Tilt, and he's gonna put in context the recent news we heard from Grayscale and BlackRock that seemed to be legitimizing the entire DeAI. ecosystem and the reason that we're doing this show. Right, Brad? Brad (01:42): Yeah, I mean everybody's talking about this report. it's been another big news topic of the week. And yeah, I mean the whole industry is seeing this as kind of validation that this giant money manager, probably the world's biggest money manager, is saying that this is the future, which is this intersection of blockchain and AI. So it's pretty pretty pretty fascinating. I mean, it's fun for us to be on the story. January Jones (02:11): Yeah, for sure. And so there were two reports that came out right together, Grayscale and BlackRock, and they were kind of both leading to to the same conclusions and and calling out some of the key people who are in this industry who are who are ones we're familiar with because they are coming in this out of the crypto and blockchain and the infrastructure side, right? And using now all these tools that is enabling AI to to to run. but that gets us back to what we're talking about today. So when we put these shows together, sometimes we're using interviews like we did today, where we had them in the past and then we're putting it together into a new narrative. And I think what we saw from the themes, the themes that are coming together in today's show is about tokens. It's the token journey. So we reference a lot of blockchain and crypto projects. And of course, we saw in crypto, you know, the token is the coin a project launches to raise money. But when we're talking about AI, there's another piece that's called a token, right? And a token is the piece of the text of the model that reads and writes. then there's the other tokens that we're talking about when you're tokenizing compute and in the markets. so did I mess that up, Brad? Or is that kind of like token, token, token? Brad (03:36): I mean, I think you got right to the spiritual soul of this discussion, which is it's so completely confusing, right? I can't can't imagine coming into this, you know, without January Jones (03:45): Ha ha ha Brad (03:49): having I I mean, I've been in covering crypto for well, geez, six years now. And so everything's token, token, token. I mean, that was the whole business model of so many of these crypto projects is just sell a token. Pump the token. so, but anyway, yeah, there's crypto tokens and then these AI tokens, which are like Four fifths of a word is kind of the way people have described it to me. On average, it's like four fifths of a word. So you can it's roughly the number of words, and but then there's input tokens, the tokens that go into the AI, and then there's the output tokens, which is what you get back. And those are also different tokens. And now the tokenizing stocks, those are really basically like crypto tokens, right? You're basically putting a stock inside a crypto tokens you can send it around on the blockchain and trade it on set exchanges and decentralized exchanges, just like cryptocurrencies. So but yeah, it's it's It must be incredibly confusing and they're totally different things, right? January Jones (05:06): It can get confusing about the language, you know, for the tokens for sure. And so we'll try to define that. but it's a good starting place. So we're gonna hear a little conversation that we had at Stanford. So as we said, we were we were there and we were at this blockchain applications conference, and so I thought that was a really good view into kind of the world that we're moving into with decentralized AI. Because it was people that were moving from the blockchain world, putting the blockchain in the back and the AI in the front, as we say. and so we talked to Steven and Gil about this, and we really wanted to hear what is happening now with projects that are coming up, coming through accelerators. How are you building a business now, even if it has blockchain in it? How are you describing it? And then also like where is the oversight and the money coming from? you remember the the conversations with Steven and Gil were about to go into those clips. What was your takeaway on on their perspective? Do you think that they've come around like they're part of the crypto remorse cycle? Of how businesses were launched for a while. Brad (06:30): Well, a lot of this was actually driven by the venture capital community out there in Silicon Valley. I mean, they would push these projects to sell a token just so they could get their money back. whereas in the old days they would take a flyer on these companies and then they would have to wait till the company went public and sold its stock. But with tokens, there was so much hype they could they can invest capital in the companies. So they're buying the equity of the companies and then somehow and then they would have these token sales. They would get all this money. The VC firms would also get an allocation of the tokens and then they would be able to sell the tokens and they would get their overall investment back a lot faster. And and so and then it just ended up killing a lot of the companies, right? If they put all their energy into managing this token and pushing the token up and that becomes their primary business. then they're not really building anything that is a useful company that would be that customers would want to use for decades. So it just sort of perverted the whole system. And I think there were some honest moments from, you know, Gil and and and Steven basically saying, yeah, the it people took it too far. And that is probably not gonna be the way that companies or that these crypto projects are making money more. So what is that new model? It's pretty interesting. I don't think it's clear exactly. January Jones (08:00): Well, let's get into a couple clips and just hear from them because really what we were talking about with Steven, we got kind of that perspective, but what we wanted to hear from him especially because he actually teaches blockchain entrepreneurship at Stanford. So we asked him what is working right now in building these businesses up and the investment space. Steven Willinger (08:22): for the most part, you can generally want to invest in a company that's gonna generate revenue soon and quickly and in large amounts. And so yeah, I think it's a great thing for the industry, right? Wh we now have founders focusing on the right thing. the number of founders I saw distracted by the wrong thing over the past couple of years is is way too high. well I mean the wrong thing is unfortunately founders, you know, react to signal just like any other, you know, being that reacts to stimulus. And founders were able to get lots of attention and raise lots of money, without checking the core boxes that a founder needs to to check, right? Which is understanding their market, building a product that delivers value and then, you know, selling and marketing that product such that they find their users and and build a profitable business. instead, they were, you know getting signal from venture investors who like, okay, like I want you to launch a token as fast as possible so that someone else can buy it. Exchanges who were, you know, similarly, it's like, okay, we we need as many tokens on our platform as possible so we can make money selling those tokens to retail investors. And we want you to spend all your time and energy marketing your token to those those same people, right? And and so that's all f fine, but if they don't ultimately use those resources to deliver something valuable, it was a lot of money wasted. Brad (09:35): I asked Steven's co founder at Blockchain Builders Fund, Gil Rosen, if raising money has gotten harder for founders in blockchain based projects. Gil Rosen (09:43): there are a few things that have shifted. One is token prices are down. Yeah. But more broadly speaking, like exit potential is also down. Right? The five to ten billion dollar FDV token launches, like those are probably not gonna happen at all, let alone anytime soon. Okay. Right? Which means that you know appetite for risk is much smaller if if your exit's gonna be five hundred to a billion dollars. So so that that's part one. Part Brad (10:05): Right, right. Gil Rosen (10:06): part two is you're only gonna have an exit if you have revenues in your real business. Now that we're able to actually use the technology and the it's viable, there's an expectation that someone is using it and you're able to like generate revenues. Okay. Right? So investors aren't willing to just underwrite technology anymore because it's not enough. Yeah. You need to underwrite there's a potential go to market, there's viability here. Okay. Right? So you know, we're trying to de-risk ourselves as well. Yeah. and and then kind of the last part is exactly what you mentioned where It is easy to vibe code a whole bunch of this stuff. It is easier to reach people. Yeah right. It is easier to you don't need as big a team to build an MVP, which you would need you know to be able to have those initial discussions. So we care a lot more about like, you know, yes, do you have the technology, but also do you have the the go-to-market distribution capability and can you at least validate that from conversations, from initial agreements? And if you can't, then there's a lot of risk there, and the people that are are able to succeed are the ones that have sector experience. And able to bring that and provide like a a distribution side to mitigate the risk of of distribution for any of these these technologies. So it is harder to I wouldn't say that it's harder to raise, I'd say the expectations are different. it's like yeah, you need to have more and yeah, you need to focus more on distribution because you need to be able to build a company, not just build a story and and a token that you can hype. Brad (11:27): What about I mean do people need less money? Yeah if they can they they do. Yeah. Yeah. Gil Rosen (11:32): A hundred percent. Like all of our teams, many of them have let go of some of their engineers, many of them just didn't end up hiring engineers. All the devs are becoming engineering managers. Yeah. And they're you know, well that's a question that we have for our teams is like, are you an AI first venture? Are you trying to optimize your functions in building your company around kind of le leveraging agentic workflows to optimize the engineering side, to optimize your go to market side and your distribution side, to optimize like all all of your you know sales interactions, your support interactions, the meetings that you're having. So it it's I think it we're doing it as a fun. We've been kind of leveraging AI to optimize a whole bunch of our work and it's been incredible and I think it's just as critical for founders. Brad (12:17): Which brings us to our feature interview. Greg Osuri has been working on this problem since 2017. He's founder and CEO of Overclock Labs, the company behind Akash Network. Akash is a marketplace where people with spare computing power rent it to people who need it. And it's known as one of the OG DeAI projects, if that's possible. We're talking about a decade. That is a long time to be working in decentralized AI, but that's the point. He was way ahead of the curve. January Jones (12:50): So first I'm gonna get into the question, it must be the money. We're gonna ask about the money. why is decentralized AI looking like a good cost alternative to what people are currently doing? Greg Osuri (13:05): Yeah. I I think we it's no longer about cost. I think it's about access and sovereignty and privacy at this point. to cost. So with enough money, sure you can get chips and enough you can buy chips, but the the exuberance of cost is like not something you would compare that, hey, a cost is like twenty percent, thirty percent cheaper. it's no longer about cost. In fact, it's right now it's just getting access to H one hundreds or or B three hundreds, rather, there is a challenge with how the contracts are designed. So you no longer have like an on-demand market anymore. The spot market is almost dead for compute. it's now moved exclusively to contracts and these inflexible opaque contracts where they average anywhere from thirty-six months to sixty month contracts with twenty-five percent to fifty percent down payment. that is inaccessible for most companies, right? So and And and and it's not getting lower, it's only getting bigger. So there are a bunch of indexes now that track you know leasing prices for B300s and H200s and H100s or whatnot. The number just keeps going up. So when the number keeps going up, it's very hard to build a business on top of it, right? For example, we ourselves, you know, Akash network, but we operate a company called Overclock Labs, which is the commercial side of the Akash network, Akash being the decentralized and open network. Overclock being the commercial you know network that, or commercial company rather, that's built on top of Overclock Labs or on top of a cash network that provides the indemnity and you know SLAs and all good stuff that you need for enterprise contracts. we also have an off-take product, which is a cash ML, which is our inference product. our inference product is how we make money on top of you know on top of Akash. We we provide managed inference for Frontier Models. Brad (15:06): An inference product again, just to sort of, you know, level set i the inference product is when you make a query of an AI and it goes and it's processing it, as opposed to the the training, right? Yeah. Greg Osuri (15:15): Correct. So when you to Chat GPT Right. So when you go to Chat GPT and ask it a question, so that that mechanism is called inference. So there are two two types of compute or two types of workloads that are run on compute, on GPUs. One is a training workload, which is, you know, compute heavy, but also it's one time and done. And inference is actually when you use AI. So now most of the demand is inference. and for us to build a meaningful business, we need price predictability. We can have an H100, you know, that was costing about $2.50 per hour. about a couple of months ago now costing about three dollar fifty cents. When the cost of goods go up so rapidly, it's impossible to build businesses on top of it, right? So it's no longer about you know just getting like I mean su yes, cost is the overwhelming Brad (16:09): Well doesn't that I mean the the ever rising costs, isn't that make that value proposition even more appealing? Greg Osuri (16:19): More compelling for decentralized computer. I think more compelling for marketplaces in general. so there are a lot of innovations that are happening with open compute. Now you can do like forwards, you can do I mean we're there's a lot of experimentation around like, hey, you know, now compute should we commoditize? Brad (16:34): These are like trading contracts and hedging and future like futures trading. Greg Osuri (16:38): Correct. Correct. So yep. So there are a lot of announcements. See CME basically announced that they're doing compute futures. there are bun the NYSE, I mean the the ICE, which is the the holding company behind NYSE, is also you know, promoting or rather indicating that they're going to do compute futures. Most of them are cash settled, which kinda is not very helpful. but Akash has been sp is a is a spot market for compute. We've been around for like six months, six years. And we are physically settled. So when you buy a computer on a cash, you're physically taking the delivery of the compute, which is actually useful for the Brad (17:14): That's pretty interesting 'cause I I mean y you know, I d it's kind of a separate topic, but I've talked with a couple of people about how these compute sharing networks might integrate with some of the financial trading around AI and and so you it sounds like you could be a you could deliver into some of these contracts if they had a physically settled, yeah. Greg Osuri (17:38): Yeah. Yeah. So Akash could become a clearinghouse because we've been Brad (17:43): huh. Greg Osuri (17:44): physically continu continuously physical settlement is how you deliver compute. Very similar to power markets, not like oil markets, right? Oil markets is you take barrels of oil delivery, right? When you buy futures when they expire, you gotta you take delivery. If you don't, you know, y or else you gotta cash settle it. Akash has been doing physical settlement, continuous physical settlement for about six years. That's what we do. You know, we we you can in a decentralized setting, it's amazing. Any party, any counterparty without identity, without sharing identity or any of the mundane stuff that you would ask for, you can freely exchange compute and a contract. So Akash is now the clearinghouse. I mean, or or else or when the f when this futures eventually has settled, Akash could become the clearinghouse. And I think that's where puck is moving and BlackRock l literally came out with a paper yesterday. January Jones (18:31): Yeah, we we covered that. Yeah. Mm-hmm. Greg Osuri (18:33): About exactly the same thing what we've been doing for about six years. So I think the January Jones (18:36): Yeah, yeah, that's good validation. Greg Osuri (18:38): thesis is coming. I mean, time is right as compute is getting more scarce here, more expensive. People want predictability, people want maturity in these markets. and financialization of compute is is inevitable. I mean, it's not gonna be like it's not a future out thing. It it is happening right now. January Jones (18:55): From your seat at the table and and longevity, how do you define decentralized AI in your work and when you talk to other people about it? Greg Osuri (19:07): Good question. So I've been in the space full time for about ten years. We launched Akash about six years ago. and I've seen quite a lot in terms of how markets moved in in this time. And decentralized AI, I think like You know, a lot of definitions as to what a decentralized AI. From from my point, it really comes down to who has control over AI and who benefits out of it. I have nothing again overwhelmingly benefits out of it, right? Is there if you consider AI to be the most impactful Full innovations in the la in the last hundred years. For comparison, AI is predicted to contribute about 3.96% of GDP by 2035 with current trends. For comparison, railroads, which were considered the biggest expansion of American economy, contributed to about 2.5% at peak. We're talking about the Gilded Era where railroads were the AI back then. Even at peak, they contributed about 2.5%. So we're talking about AI more than anything we experienced in terms of GDP contribution in the history of America. So, you know, of course, I don't expect people to remember how Gilded Era was, it was, but it was crazy. Gilded era was is what led to the roaring twenties and a whole lot. Things and that's very similar, you know, things that are happening right now in AI. And there is enormous regulatory you know arbitraging by bigger companies. We saw you know Anthropic and open AI, even Elon Musk's xAI come up with recent threats about like potentially taking out humanity by 2030. all these you know scare tactics are not Uncommon. We saw this in the Gilded era. If you study the study history, you'll know ex what these railroad companies were doing. And same tech tactics are applying now. So if you have AI controlled by a few elites, few oligarchs, we're going to end up like railroads. That's the reality. And railroads did not take off in America, not because they were we were not technologically advanced or we didn't have the intention. It's just they got so oligopolized and so powerful politicized that they started they they're the very definition of crony capitalism, right? That's this the they began this idea of crony capitalism where you had regulatory advantage and only big companies could have the advantage and smaller guys could not and and a whole lot of things that led to regulation in America. We didn't have regulation before railroads, right? The idea of regulation came because of railroads, right? Because you know the January Jones (22:06): Because of this control over basically the whole economy of the country at at the time, right? Because even though it was a a lower percentage overall of what you're saying, it was really running what was happening for jobs and expansion and and all kinds of resources, mining especially, right? yeah. Greg Osuri (22:25): Yeah, everything ran on railroads. So you c you mean railroads ended up being real est realistic real estate trusts. They would they would have crony they would have politicians that would, you know, bribe them and get all kinds of favors. When you have too much power. Brad (22:40): going back to this idea of, you know, you ha you had these these you know, tycoons and they're controlling a small set of companies and you know, and maybe coordinating with each other in some way. And then I mean in a good market, in a good healthy market, competition is the antidote, right? And so Greg Osuri (23:04): Okay. Brad (23:05): here you have these big centralized AI firms and they are making incredible advances. I mean we see it every day, just like the f the models get better and better. And I was just reading the models are actually Getting cheaper too. The AI, it gets a lot cheaper every quarter for the same performance. but is, I mean, going back to decentralized AI, is that the competition? I was there was also a similar article in in the Wall Street Journal about electricity, and when that came about, and Edison was pushing D C you know, direct current DC, and then Westinghouse came out with AC, and Edison was talking about all the dangers of AC, but The reality was there was an alternative there. And I was just wondering if that is decentralized AI versus AI or you Greg Osuri (23:55): Correct. So Brad (23:56): know. Yeah. January Jones (23:56): that decentralized AI is the D C to the A C? Brad (24:01): Well, it's the A C Greg Osuri (24:02): AIST PUDI. Brad (24:02): to the D C But I don't know. What do you think about that, Greg? January Jones (24:04): it's the A C D C D C C Greg Osuri (24:07): Yeah, exactly. it's a great I was. January Jones (24:10): What's your question again, Brad? Could you phrase that again? Brad (24:12): Well, the question is, you know, when you're talking about a a small group of companies controlling, you know, the means of production or, you know, the the valuable capitalist enterprise and and then you know somebody else and they can tr they also control the prices, right? And they get all the revenue. And so then if you have competition an alternative way, maybe that provides a balances the market. So th in that way would decentralized data potentially help to reduce prices. Greg Osuri (24:45): I it's it's happening right now. so right now the way to use AI, there are two places you can get AI from, broadly speaking. One is centralized companies, Chat GPT, Anthropic, and O X X AI, and next another one is open source models. GLM five, Deep Seek four flash four four one, Qwen family of models. they're incredible open source models that do really well, they come very close, if not Better in some cases than the centralized model. Well, open source models are free to use as long as you have a computer. Closed models are not free to use, you gotta pay them, right? So every time you use an open source model, that means less money to these companies. So they're going to do a everything in their power to make sure that you use their models because there's more money for them. It's very simple, right? The value incentives are very well aligned. As an AI user, I want to pay less money, right? I want to spend less money because it's very, very expensive. in order for me to do that, I will either go find cheap GPUs or buy GPUs if I can, which is also getting very hard because the demand is so high right now. or ideally, if I'm a gamer, I would use my existing GPUs. because a lot of these gaming PCs are getting very, very good. Like very good. I mean, I I run my own home, a A Qwen 3.6 on 5090s. Qwen 3.6 is a an incredibly capable model. It is a small model, it is not frontier, but it can do most of the work in my home, like monitoring, like video feeds, audio feeds, and a whole lot of things that I want to automate my home. I don't that I don't need to use Chat GPT. you know, we is my cost is significantly lower than most people's costs. That's because I decentralized AI, right? Because I can have local AI. And now we have a program where I'm not using my AI all the time because it's just my family usage. It it it's useful in I'm maybe using 20-30% of the time, but there's 70% of time I'm not using my own local chips that I give back to Akash Network. So I'm actually making January Jones (26:56): Okay, well let's get into that. So Greg Osuri (26:58): money running my own AI. I'm getting my AI for free and making money to buy more chips. January Jones (27:05): Well well, you were talking about this when we were kind of warming up for the interview, but I find this quite fascinating. And so we are talking about costs. Brad was was framing it as a cost alternative to centralized AI and and having more control, but you also have your own really interesting theory that maybe people haven't thought about, which is that your home, you can be your own data center. Basically. Greg Osuri (27:32): Mm-hmm. January Jones (27:32): Tell me about that thesis, how that's built into Overclock and and Akash and and kind of what you do in this space. Greg Osuri (27:40): I would propose rebranding data centers to token factories. January Jones (27:45): Okay. Greg Osuri (27:46): For you know. Brad (27:48): Refinery. They're like refineries, right? I mean yeah. Greg Osuri (27:50): Token refineries, token factories. So I'd run a token factory at home, which makes me tokens, right? Or else I had to go buy these tokens from centralized companies, which I don't want to. I it's like growing my own food, right? I mean, I'd rather grow my own food if I have time and energy and patience for it, because I know it's cleaner, it's better, cheaper. I mean, this is very straightforward. And if I if there's a way for me to sell my excess food, I can even make more money. It's the same concept, right? It's like you don't have to get your food from a from a distributor or a or a or a a a grocery store which you n may not know what the where the food came from, right? there's all kinds of things now. Like it's getting scarier, in fact, like to get to to get even asparagus from from from stores. So the good thing about having local AI is you know you have privacy. You have you know you have because it's not leaving your own network and it's cheap. It's n it doesn't cost you anything. And you can make money selling that excess compute or capacity on Akash Network. And there's so many others now that you can actually sell inference directly, right? So many other marketplaces. Open Router is a great marketplace that came out and Stripe now owns them and Akash ML is one of the providers of OpenRouter. And Vercel has them, you know, you have Hugging Face as as a as a token marketplaces now. So there's a lot of marketplaces that are popping up and their capacity is just growing. In a in a crazy manner. That's why Stripe paid seven billion dollars for a open marketplace for inference. So inference demand is not slowing. If you have compute, you can sell that right now and you can make a lot of money. We predict if you to Akash Network, I Akash ML. Akash Home Node, that's home node. It is a program you can download directly from your for your on your Windows PC that you're playing your games on, and it'll start making money from day one. January Jones (29:39): So you're talking about, and I and the hint of this it did come through like people that have like bigger systems at home. And I come that as like, man, the gaming computer my son built and I had to pay for was a monster. I'm like, what are you doing with this? Like, and why do we have to see into it? And why are there neon lights Greg Osuri (29:55): Yeah. Exactly. Brad (29:57): Same. I got a few of in in this house, yeah. January Jones (30:02): in it? And like, what is going on with this monster? But coming from that perspective, Brad (30:06): Cool keyboard. January Jones (30:09): Is that you're saying when people beef up their own connection at home, that there's a way for people to really be owning their own AI compute and then selling it? Greg Osuri (30:26): Exactly. Ex you can do that right now on Akash, like home note. Akash Network. there's also a calculator as to it'll show you how much you can make. I mean we tend to be on the lower we tend to be a little more conservative on what we promise. People actually make a lot more than what the calculator shows. Like right now, fifty ninety is a going for like sixty five cents. So that's like, you know, I don't do the math, right? Like if you're giving, you know, twenty four hours, let's see. That's about like fifteen dollars per day and that's about f close to four fifty dollars a month. And add that up and you can actually recover all your costs that you spent building the gaming machine within a year. Like January Jones (31:06): See, he wasn't enterprising like that, my son. right. Brad (31:09): Ha ha Greg Osuri (31:09): And you can that means you can buy more games and more GPUs. You know, think about it. Like, what is five hundred dollars for you know, I don't know how old your kid is, teenager. Brad (31:18): Well, I mean if you think about January Jones (31:19): Okay, teenagers of the world, Brad (31:20): it January Jones (31:21): teenagers of the world, make your parents proud. Brad (31:24): That's your new market, Greg. Greg Osuri (31:27): I know. Brad (31:29): But I mean, if you think about it, like you January Jones (31:28): Seriously, we've invested in you. Now do something Brad (31:32): you could have solar panels and I I mean I have s some solar panels and not nearly enough to run my house, but it's like a credit on my bill. But so I'm basically selling electricity to the electric Greg Osuri (31:44): Yes. Brad (31:46): company. But you could just keep the electricity January Jones (31:48): So it's the same. Brad (31:49): and sell the compute. Theoretically, right? If really? Greg Osuri (31:52): So much better. Compute. I have this thesis called compute as a load balancer for energy. It's way better the load balancer for energy than you than your utility companies, because you make so much more per watt, per kilowatt, January Jones (32:07): Mm. Greg Osuri (32:08): putting compute and abstracting energy in compute, then you would sell back to the grid. Grid, I don't know how much grid pays you, but in California they pay like two or three cents, which is per per kilowatt hour, which is it's it's it's ridiculous, right? I could Take that two or three cents, give it to my AI, and I can make 61 cents per hour. If I have my CapEx already covered, which you know, which you already bought, a gaming PC, right? And a gaming PC, we're not talking about specialist hardware. If you put specialized hardware, I would do check Pro 6000s, for example, which is in high demand, which costs about, I don't know, used to cost seven grand a pop. a Pro 6000s now will make about $2 per hour for that two cents you're getting, giving back to the utility company. Like right now. January Jones (32:51): Mm, that's a third of minimum wage. Federal minimum wage. Greg Osuri (32:56): Correct. January Jones (32:56): Nuts. Greg Osuri (32:57): Along with you getting free AI. And that's the best part. Like not paying January Jones (33:01): Yeah. Greg Osuri (33:02): AI companies. January Jones (33:03): Yeah, I mean that is a different way to frame it in in showing where you do have control, because I think that is an issue with the public's perception of AI. it seems like the government can't control it, even if the companies are asking for it. and that you feel like, Well, what can I do? There's no there's no kind of way to have my own sovereignty and to figure out how I can keep Using technology, keep using AI, but also maybe stick to my values a little bit, right? And not really compromise about everything that I've I've kind of like lived my life by in the technology and the kinds of communities I've wanted to build. So let's frame decentralized AI like that. Because in all the things you're saying, you have a way of thinking about the world in a decentralized frame. So tell me about how. that r is reflected in a caution, but also the bigger picture of how that's influences your worldview that may be more relatable for some of our audience. Greg Osuri (34:09): Yeah, when I come like when I say decentralized AI, when I say control, it all comes down to like as technology matures, it gets commoditized, right? Cloud, for example, as it matures, it got commoditized. It became so ubiquitous and so necessary for all our lives. We pay a cloud tax without even realizing we pay a cloud tax. Like Amazon, like cloud companies, Amazon being the biggest one, you know, I think every half of every dollar you spend on online services goes to the And you have no say in that that that tax you pay. similarly, I mean that cloud became AI now because most of the cloud companies or hyperscalers are AI companies now, right? AI compute providers. it just metamorphed into this new thing, and it's even bigger problem now, right? So where is the incentive? Where's the money going and who is that benefiting? Now where's the money coming from? Well, it's cheap debt. January Jones (35:00): That's the questions. And is Greg Osuri (35:02): Yeah. January Jones (35:02): it supplemented, right? There's so many costs that you don't see. Greg Osuri (35:06): Yeah, yeah, yeah. Brad (35:07): Well also, is that going to lead to massive overcapacity? Greg Osuri (35:12): Not necessarily. I don't think we have capacity nearly to to address the need, but I think it's going to lead to a massive demand for all things, you know, that we depend on. Energy, for example. PJM is the biggest grid in America. It serves about 13 states, right? It's the state it's the grid that serves Virginia, for example. The energy prices in Virginia, which is the biggest like data center like concentrated place, went up by five times. And who do you think is paying the cost? People. Like normal January Jones (35:44): yeah. Greg Osuri (35:45): people. Five times over the last six years. And that's a crop. January Jones (35:48): Yeah, it's supplemented. Your your energy bill goes up and that's what's got people so upset. So so reframe Greg Osuri (35:52): Energy Exactly. January Jones (35:54): this for me though. Like if you're mad about that, that's happening to you, that's when you first start noticing what is this AI data center thing happening, right? Greg Osuri (36:01): Yeah. Mm-hmm. January Jones (36:03): And so tell me, like, what does decentralized AI in a cosh network, how's the the pivot of thinking and access, Greg Osuri (36:11): Right. So in that scenario where your AI bill is going up so much, what we are proposing is instead of building new data centers, let's go to houses, let's go to homes, let's go to places where there's abundance of energy that comes from sun or wind or whatever sources you can get, and let the people benefit directly, instead of just like, sure, AI is benefiting humanity, I agree, but not directly benefiting the people that are paying the price for AI. Which are normal people, either through energy, through water bills, whatever. I mean, I think the water thing is the whole meme, but the point is it's getting impacted heavily, right? And the people can benefit significantly by using their own AI. And they're not able to use AI because AI is very expensive. I mean, AI, you cannot afford agents. I mean, normal people cannot afford agents unless you have a strong business case. Like we spend so much money on AI, on on centralized AI because you know, we still use a good mix of Yeah, models, because we have no choice. And we are almost spending as much. I mean, we used to at one point spend as much money as to as we would spend on an engineer to pay the salaries on a model for an engineer. Like the same same pay. And that's not affordable by most people, right? So if you want AI to benefit people, the cost has to come lower. And the way to bring costs lower is to lower the energy cost and access costs. And the way we do that is by moving. AI to weigh the compute to where the energy is instead of trying to move AI to instead of instead of trying to move compute to where the energy opposite, instead of moving energy to where the compute is, right? And how we think about AI Akash Network is very straightforward. We're we've we've you know we've always looked at commodities that become utilities essentially, water, air, energy, compute. January Jones (38:02): Yeah. Greg Osuri (38:03): so you know Brad (38:03): Well, if I can just try a different angle, capitalism, that's our system currently, and we have we see these giant AI firms, they're getting so much money, right? I mean They're getting revenue. They you know, people are signing up for for lots of subscriptions from them. on the other hand, it's not totally clear what their costs are versus the price they're selling it for, what their profit is. But I'm just wondering, you know, decentralized AI, does it it doesn't seem based on what we're seeing in the markets. It doesn't seem like there's a huge, you know, trade on in DeAI tokens or equity or whatever there is to invest in DeAI. I'm just curious like, how do you compete against that capital machine? Greg Osuri (39:01): So it's DeAI is not a capital problem, in my opinion. It's not I I don't think it's it's under or overcapitalized. It is a knowledge problem. Like the beauty of DeAI or decentralized systems in general is it's not money, it's the will of the people, and that's beauty. That's the capital for decentralized AI. If today ten thousand nodes join Akash network because they all see the value in providing their nodes, that is that'll make Akash one of the biggest networks, right? How about hundred thousand? How about a hundred million? There are definitely a million PCs in the world that can join the network. So when you have so much supply coming in, I mean that's gonna attract, you know, attract the the the the companies that wanna use the supply. Right? So it's not it's not a it's not a capital problem. I don't Brad (39:49): So it's just the market hasn't figured that out yet, is kinda what you're saying. Okay. Okay. Greg Osuri (39:53): Exactly. It's very underpriced, very under under like literally, I mean like just yesterday BlackRock came and said, Hey, compute should be tokenized. Well, that's been our thesis for six years. I feel I feel vindicated. I feel incredible, right? January Jones (40:06): Mm-hmm. Yeah, how do you feel about that? Black rock coming in. You feel vindicated, yeah. Greg Osuri (40:12): In fact, I c commented asking like, hey, which is the most mature network that can deliver this what BlackRock is proposing? Came with Akash. I mean to Graph, yeah, you know, like you know, Grok came out with an answer. January Jones (40:18): Ha you're like, look over here. Brad (40:20): I think you were me you were Akash was actually mentioned in that report, I believe. Greg Osuri (40:25): In the Grayscale report I think was mentioned, but not the Black Rock. I th I don't think BlackRock report mentioned any companies, but sure, they mentioned Brad (40:27): the gr Okay. Okay. Greg Osuri (40:31): e our our our white paper, basically, right? Our own pieces, right? Like Brad (40:34): Maybe I was thinking of the Grayscale paper, but I yeah. Okay. Greg Osuri (40:38): Right. Grayscale did mention about Akash. Grayscale did a wonderful job understanding, breaking down the structural problems. Brad (40:44): I mean they have bags though, right? I mean they're they're i you know, they're an investor in the space. versus BlackRock Greg Osuri (40:50): Right. But Brad (40:53): is you know, is just a giant you know, they're they're big enough to be d Greg Osuri (40:56): It is the biggest asset manager. Brad (40:57): they're big enough to not be opinionated Greg Osuri (41:00): They're the biggest asset manager in the world. I mean, there's no doubt about it. It's not it's not TC's and I felt T C's just caught up structural problem. Right. January Jones (41:08): That's our thesis too. That's why we're doing this show, right? Because like we're seeing everybody start to pay attention to this space. And that's why it's so great to have your perspective from someone that's been in this space, January Jones (41:20): In our endeavors to explain and understand things, like, yeah, how does this shake out at Wall Street? If we're in the beginning of a market being made, who's getting in there? How are we, you know, deciphering this information? and so today for the news roundup, I wanted to talk to someone who reads market signals for a living. So I talked to Ivo Enchev and he leads legal and product at Tilt and they build investment indexes. So I wanted to ask his take on the reports from BlackRock and Grayscale and what this really means for what's to come. January Jones (41:59): thank you for being here, Ivo, and joining us on the DeAI news roundup. Ivo Entchev (42:04): My pleasure, thanks for having me. January Jones (42:07): So you are the perfect person, I think, that I've met in the crypto space over the last few years to talk about this story in particular. two of the biggest names in investing, Grayscale and BlackRock, both put out reports last week. talking about that the decentralized AI market, which we see as kind of a very broad space, really has some teeth now. They're seeing compute being such a game changer. And in fact, the same day Grayscale came out with that report, they renamed its Bitcoin miners ETF the AI Compute ETF. The BlackRock report is saying that AI agents will become crypto's next big customer. And we've been reporting about that at DeAI News. this is all about kind of seeing where everything's going. And so let's get your perspective on this. Grayscale and BlackRock, you come from a venture capital background. You're also a lawyer. This is kind of a perfect mix for market signals. Tell me how you're reading this. Ivo Entchev (43:22): I think you're absolutely right. that this was a big week in the area of kind of you know crypto and AI. I think it's a moment that we've been waiting for for a while to see a clear path forward for the two technologies converging. you know, we often talk about AI disrupting industries because of the intelligence that comes in and and the efficiencies that are able to be gained through the use of that intelligence. but you know what what maybe people or at least ordinary people using these products don't fully understand is that underneath all of that is really a new unit of account, right? A new form of, I guess, oil or commodity for this intelligence economy, right, in the form of of January Jones (44:08): Mm, mm. Yeah. Ivo Entchev (44:09): compute and tokens. And isn't it I you know almost ironic, but certainly fitting that you know the industry that the pre-existing industry of crypto that denominates itself in tokens is the one to actually tokenize intelligence tokens, AI tokens, you know, in this new agentic paradigm. so I think what we saw in in both reports that you mentioned is big players identifying that we have a new asset class. that we might just call, you know, compute or a t a compute commodity that will need to be traded rapidly in an agentic economy that requires you know agents to to fund to accomplish goals i in a kind of in a native format that's not you know using credit cards or invoicing users. January Jones (45:08): Yeah, well the way you're setting that up, to me I'm thinking, okay, we have to give agents machine money so they can trade and process things fast enough to pay for their own compute. Like what is happening here? Ivo Entchev (45:28): Yeah, I I think that's essentially it. as as agents become more prevalent, I think we're going to recognize that you know, existing institutions and infrastructures that we have are just not adequate. you know, that we need, you know, machine readable, machine accessible, swift rails for these agents to accomplish their goals. and you know, these large institutions who put out the reports I think are looking ahead and recognizing that the most efficient way for the agent economy to work i is for agents and themselves to to acquire compute tokens to trade them to fund themselves and to to deploy them you know th they're visions of this future that are you know depending on where you sit exciting or you know scary, but you can imagine a point when you'll be able to literally invest in an agent and that agent's activities online and any return that agent is is is able to to achieve and it certainly is part of that process it will need to trade its own compute and we're already seeing the beginnings of that asset class emerging and it's very likely to grow and and and and really become one of the dominant asset classes in the world. January Jones (47:02): Yeah, I think one of the things that's interesting is to the timing. And so we're having kind of this validation. on the show earlier. we talked to Greg from a Kosh network. He's feeling himself. He got named in that Grayscale report. he's like, Yeah, I told you guys I've been on that, right? about this decentralized compute and the markets and and the power needed. Ivo Entchev (47:29): Well, i i first of all I do think it's emboldening to to anyone in the space that who has been anticipating this moment of intersection and has been identifying this asset class as one that should appear natively on chain and be accessible to agents. And as you pointed out, I think the space was was quite early with some of these deep end projects focusing specifically on compute and and being able to aggregate compute through you know decentralized networks and and supply where it was needed through through a completely different business model, I think when we talk about the maturation of the asset class, right, there are priors. Like we can look at Bitcoin, I think, as an example of a n of a completely new asset class, right? Did not exist. emerging asset, as those who bet early will tell you, you know, one that with enormous upside that that was invested in by by true believers at the outset, but required a lot of intermediate kind of you know, successes, right? and and ecosystem partners to to develop in order to to lead it to its ultimate sort of triumph, you know, is remarkably a a j a a a government sanctioned, you know, digital asset, and January Jones (48:55): Yeah. Ivo Entchev (48:56): sanctioned in a good way, right? but you know, there's a process of maturation, right, that these these assets go through this that that starts with, you know, spot trading. that, you know, as those will remember who were in the space early on, is is often dislocated, meaning between venues. You know, there were people who did quite well in the early days of Bitcoin, for example, just arbitraging the price of Bitcoin across crypto exchanges. You know, that that was just the very nascent stage. And then you have of course, you know, price infrastructure develop. You have this arbitrage opportunities being eroded. and you know futures markets and then eventually etfs and and and these sorts of things and I suspect we'll see the exact same path here arguably on an accelerated basis for compute commodities. What I found interesting just because I work currently more in the index space is that we're seeing now the emergence of indices, price indices for AI compute. and and people are interested in knowing what what is the price of compute at any given time and they consult indices for that information. that to me is is just another indication that we are very fast looking ahead to the maturation of this of this asset class into something that you might hold in your ETF eventually. January Jones (50:19): Well, thank you so much for joining us today on the DeAI News Roundup. Ivo Entchev (50:24): My pleasure. Thank you for having me. January Jones (50:25): Alright, well, thank you so much for joining us. check out our stories. We're publishing all the time on DeAINews.com. we're on socials. We have a TikTok, even though we're old. we're also doing X and LinkedIn and YouTube. And we're really posting great clips from all our guests. And you could also find extended interviews and hear more from them than what we do produce in the show. So please check us out there, and try to learn some more about what we're doing. I'm January Jones. Thanks for being with us here on the DeAI News Journal. Brad (51:04): And I'm Brad Keoun. ``` --- title: "CNBC highlights BitMEX founder's bet that AI bust will boost Bitcoin" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-07T14:16:43.251Z canonical: https://app.distro.media/newslode/deai-news/cnbc-highlights-bitmex-founder-s-bet-that-ai-bust-will-boost --- # CNBC highlights BitMEX founder's bet that AI bust will boost Bitcoin Arthur Hayes, the crypto-exchange founder who now runs his own investment company, Maelstrom, reportedly said at a conference in Singapore that the trillions of dollars being spent on AI data centers will leave the industry overbuilt, as reported by CNBC. The result will be excess capacity of computation, a market crash and ultimately a bailout that will strengthen the case for buying Bitcoin, Hayes said. **The DeAI angle:** Hayes has made similar comments before, and his new venture, Flop, aims to build a payments network letting AI agents convert tokens directly into computation power. What's noteworthy about the CNBC article is that the DeAI-oriented thesis is now getting spread in the mainstream financial press. The Hayes remarks also garnered coverage from [Yahoo Finance](https://finance.yahoo.com/markets/crypto/articles/arthur-hayes-expects-ai-boom-035312951.html) and [Quartz](https://qz.com/arthur-hayes-bitmex-ai-data-center-crash-bitcoin-100726). --- title: "Sony-backed Soneium says AI agents could license music on-chain" author: "Bradley Keoun" authors: ["Bradley Keoun"] publication: "DeAI News" date: 2026-10-06T23:25:25.735Z canonical: https://app.distro.media/newslode/deai-news/sony-backed-soneium-says-ai-agents-could-license-music-on-ch --- # Sony-backed Soneium says AI agents could license music on-chain Sony-backed blockchain Soneium on Tuesday said it plans to support ERC-8432, a proposed Ethereum standard designed to let people and AI agents discover creative works, read licensing terms and record agreements on-chain. A rights holder could register a song and publish machine-readable terms, according to a [Soneium blog post](https://soneium.org/en/blog/erc-8432/). An AI agent making a video could then find the recording, check whether it can license the work and acquire an agreement in one transaction. The video app could verify that agreement on-chain before using the track. The proposal targets a gap between owning tokenized intellectual property and knowing what may legally be done with it. [ERC-8432](https://ethereum-magicians.org/t/erc-8432-onchain-ip-asset-and-license-registry/29792) would give independent registries a common interface and identify each record by its blockchain, registry and ID, rather than requiring one central catalog. Soneium is an Ethereum layer-2 network developed by Sony Block Solutions Labs, a joint venture between Sony Group and Startale Group. Its [mainnet launched in January 2025](https://soneium.org/en/blog/soneium-mainnet-is-live), and the project has emphasized entertainment and creator applications. Since it's an Ethereum layer-2 chain, it adheres to the same standards as Ethereum, which is the world's largest programmable blockchain. **The DeAI angle:** Blockchain projects are rapidly working to develop standards that would make it easier for creators and their AI agents to share and use intellectual property rights, in a copyright-respecting way that includes a mechanism for monetization. * * * _HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: I created the first version of this piece on Hermes using an OpenAI model and a custom skill designed to write drafts based on blog posts. I then filed the draft directly to DeAI News on DistroVerse, our web app, where I used the built-in story-editing interface to do a final edit. I created the story illustration on Hermes with the OpenAI model and a custom story-illustration skill. I checked the facts prior to publication._ --- title: "Crypto AI sector surges 54% in September, Grayscale says" author: "DeAI News staff" authors: ["DeAI News staff"] publication: "DeAI News" date: 2026-10-06T16:14:49.007Z canonical: https://app.distro.media/newslode/deai-news/crypto-ai-sector-surges-54-in-september-grayscale-says --- # Crypto AI sector surges 54% in September, Grayscale says Grayscale reported that the crypto AI sector gained 54% in September, double the broader crypto market's return, as reported by CoinPedia's Yash Jain. The move left the sector's value at roughly $15 billion, according to the report. NEAR Protocol's token jumped 183%, while Bittensor rose 37%, World gained 47%, and Venice climbed 70%, according to Grayscale.