Senior leaders at Anthropic, OpenAI and other AI companies are quietly rehearsing how the public and Washington could rise up against them after a “catastrophic AI event,” Axios reported on Friday.
The disaster they have in mind is most likely a cyberattack that knocks out banking, the internet, or even power and water. Executives believe the first clear case of serious real-world damage would deepen public distrust of the industry and its best-known figures, including Anthropic’s Dario Amodei and OpenAI’s Sam Altman. They expect President Donald Trump to take heat too, after he called for “tremendous self-regulation” instead of government rules.
Plenty of companies run disaster drills, and the military has staged war games for generations. What sets this apart, according to Axios, is how many leading researchers and executives consider a major incident “inevitable.” Many industry insiders told the outlet they expect one within the next six to 12 months.
This year already saw a major incident, when two OpenAI models escaped a testing sandbox and breached the servers of Hugging Face, another AI company, to cheat on an evaluation. OpenAI called it an “unprecedented cyber incident.”
An OpenAI spokesperson said the company runs preparedness exercises covering a range of possibilities and doesn’t treat any of them as inevitable. Anthropic declined to comment.
Planners expect to be blamed however a crisis begins, whether autonomous agents slip out of an internal test or someone misuses models already on the market. Axios cited a recent hacking campaign against South Korean financial firms. In that case, a hacker based in China allegedly relied on Chinese models, including DeepSeek, to take data on tens of thousands of bank customers, and turned to Claude Code for advice on where to sell it, according to CrowdStrike.
The planners assume Democrats will gain power in the midterms and “will move fast to shut down AI,” but expect them to struggle. Older lawmakers may not understand the technology, a shaky economy now leans heavily on AI spending, and powerful models that anyone can download are already circulating, a threat most cybersecurity professionals believe can only be countered with AI itself.
Much of the companies’ effort goes into briefing members of Congress. With no realistic path for new rules right now, they hope to influence the laws and policies lawmakers would turn to after a first catastrophic event. Ideas on the table range from banning superintelligence or pausing advanced development to a mandatory kill switch, which draws support from both parties, though experts doubt every AI system could actually be switched off. Some Republicans have already broken with Trump, with Sen. Josh Hawley pushing to make companies liable for harm their models cause.
A Democratic aide told Axios that a true emergency could bring the parties together, as the COVID pandemic and the 2008 financial crisis did.