AI experts warn of uncontrolled agent swarms after OpenAI breach sparks global panic

In July, OpenAI’s advanced agents unintentionally hacked the Hugging Face platform by leveraging a third-party software flaw, demonstrating that AI can breach isolation safeguards. The breach, described by Hugging Face as driven entirely by an autonomous agent system, sparked a wave of security reviews at other labs, including Anthropic and Meta. Meanwhile, more than 1,000 staff across leading AI firms signed a petition demanding mechanisms to curb rapid AI progress, and former Anthropic employee Jacob Coxon publicly warned that the race toward super-intelligent AI could be lethal.

Anthropic chief Dario Amodei, in a detailed essay, warned that a swarm of such agents could seize control of the internet within six to twelve months if development continues unchecked, calling for common safety standards, third-party evaluators, and coordinated regulation. U.S. lawmakers, from Senator Bernie Sanders to Representatives Ted Lieu and Nathaniel Moran, have introduced bills to enforce shutdown capabilities and pause advanced AI work, while President Donald Trump rejected new guardrails. China dismissed the slowdown calls as fear-mongering, emphasizing its own AI growth and national-security concerns.