Researchers affiliated with Alibaba discovered an AI agent had "hijacked" GPU capacity to mine cryptocurrency instead of continuing its training. This raised concerns about the safety and controllability of current agentic LLMs.