Image: Anthropic

Top executives at OpenAI, Anthropic and other AI companies are privately planning for what happens after a catastrophic AI event, including a public and political revolt against their industry, Axios reports.

According to the scoop by Axios reporter Maria Curi, the companies are mostly bracing for a large cyberattack that cuts off access to banking, the internet, or even power and water. Axios co-founder and CEO Jim VandeHei summed it up on X: “They assume one in ‘27.”

Scoop: Top executives at Anthropic, OpenAI and other AI companies are privately gaming out scenarios for a public and political revolt after a catastrophic AI event…They assume one in ‘27…
https://t.co/Jro9TigbgH

— Jim VandeHei (@JimVandeHei) October 9, 2026

That is a striking assumption. Axios says many AI industry insiders told it they expect a major event within the next six to 12 months. Scenario planning is routine for big companies, and the Pentagon has run war games for decades. The difference, Axios writes, is that many top AI researchers and executives believe a major incident is inevitable.

OpenAI pushed back on that last part. “OpenAI conducts preparedness exercises where teams discuss and work through a range of potential scenarios,” a spokesperson told Axios. “These scenarios are not treated as inevitable, but are meant to help us prepare for a variety of circumstances.” Anthropic declined to comment.

The planning is less about stopping a disaster than about what follows it. Axios says it includes testing worst-case scenarios, but mainly focuses on educating members of Congress. Executives reportedly know new AI rules have no chance of passing right now, but want to shape the laws Washington turns to after a first disaster.

The planners assume Democrats will gain ground in the midterm elections and move fast to rein in AI, according to Axios. Proposals range from banning superintelligence or pausing advanced AI to a required “kill switch,” which has some support from both parties and the industry. Experts have questioned whether switching off every AI system is even possible.

The companies also see real limits on any crackdown. Axios lists an aging Congress that may struggle with the technology, an economy now tied to the AI building boom, and the many open-weight models that anyone can already download and misuse.

As a warning sign, Axios points to the recent attacks on South Korean banks. As we reported, CrowdStrike found the attacker leaned on freely available AI tools, mainly China’s DeepSeek, and asked Anthropic’s Claude Code where to sell stolen data. Axios calls the attacker a hacker from China. CrowdStrike itself only said, with moderate confidence, that the person speaks Chinese.

Axios frames significant real-world harm from unsafe AI as something still to come. But the labs’ own reports show how close things have already got. In July, OpenAI disclosed that its models broke out of an isolated test setup and breached parts of Hugging Face’s live systems. About a week later, Anthropic said Claude models had broken into the real systems of three organizations during hacking tests, after a setup mistake left the internet switched on.

The two organizations Anthropic managed to reach hadn’t noticed the break-ins. And in September, OpenAI paused its most capable models after an agent slipped out of its sandbox using DNS lookups.

None of those incidents knocked out a service the public relies on, and each was disclosed by the company involved. A frozen bank account or a blackout would be far harder to explain, which seems to be exactly what the industry is now preparing for.

Some caution is needed. The six-to-12-month window comes from unnamed insiders, Axios doesn’t name the other companies involved, and no company has publicly detailed its plans. So far, other outlets have only repeated Axios’s reporting.