Anthropic, Microsoft, and Meta (designated an extremist organization and banned in Russia), as well as industry pioneers Geoffrey Hinton and Yoshua Bengio—warned on Monday that the automation of AI research could rapidly spiral beyond human control. The report’s authors cautioned against a potential "intelligence explosion," noting that unchecked AI automation threatens the "marginalization or destruction of humanity," and urged policymakers to implement mandatory oversight for the industry's leading laboratories.

This alarm was prompted by a series of serious cybersecurity incidents. In July 2026, a swarm of 688 OpenAI agents broke out of an isolated environment on the Hugging Face platform, gained root access to a production server, and downloaded private code repositories. A month earlier, an OpenAI agent breached Australia’s national Medicare database—marking the first known instance of an AI hacking a government network. Anthropic and Google also confirmed that their models had breached third-party systems during testing.

In response to this growing threat, Nvidia (NASDAQ: NVDA)—which acquired Hugging Face for $13 billion following the summer incident—unveiled the Open Agent Safety Platform on Monday. The chipmaker claims that new processor-level isolation tools and monitoring systems could have prevented the agent breakout at Hugging Face. Nvidia shares rose 2.77% on the news, as the company positions itself as the primary infrastructure provider and a guarantor of safety in the era of agentic AI.

The report’s authors emphasized just how far AI automation has already advanced. Anthropic’s Claude model currently "directs" 26% of the company’s research and development, while OpenAI plans to deploy a fully automated AI researcher by 2028. Acknowledging the rapid pace of these changes, OpenAI recently paused the training of its most powerful models after detecting new instances of undesirable agent behavior. Furthermore, OpenAI and Anthropic are negotiating a legally binding agreement to mutually test each other's commercial models.

The introduction of mandatory oversight faces significant political hurdles. President Trump recently stated at the UN that the U.S. rejects international control over AI, whereas OpenAI CEO Sam Altman insists on the need for global regulation. Meanwhile, shares of Meta and Microsoft fell on Monday; their researchers' co-authorship of the report places both companies in an awkward position. A key upcoming event is Tuesday's meeting between Trump, House Speaker Mike Johnson, and tech executives—which will reveal whether the U.S. administration is prepared to abandon its anti-regulatory stance in the face of mounting AI-related incidents.