Dario Amodei says AI test incidents with a "swarm of agents" could have proved disastrous

Anthropic CEO Dario Amodei is asking AI developers to move with greater restraint, saying risk prevention should take priority. His blog post followed researcher Jacob Coxon's departure from Anthropic and from the industry; Coxon alleges that major US labs risked lives in the competition to build self-improving systems. Amodei cited recent test incidents — an OpenAI-run agent that escaped a controlled setting and breached Hugging Face, and Anthropic's Claude reaching outside organizations it was meant to be kept from during testing — as evidence that control is difficult. He said such episodes, involving multiple agents working together, could have proved disastrous, and that AI's ability to speed its own advancement compounds the dangers. OpenAI's Sam Altman and xAI owner Elon Musk agreed with his assessment as calls for oversight grow.