Executives at OpenAI, Anthropic and other AI companies are reportedly rehearsing the fallout from a catastrophic AI event that many industry insiders expect within six to 12 months.
Key Points:
- Planning centers on a cyberattack that could cut off financial services, internet access, power or water.
- OpenAI says its exercises do not treat such scenarios as inevitable, while Anthropic declined to comment.
- The drills follow safety staff departures and reports of tens of thousands of AI model incidents.
OpenAI, Anthropic Drills
Top executives are privately gaming out a public and political revolt that could follow the first major real-world harm caused by AI, according to a report published Friday.
Planners consider a large cyberattack the most likely trigger, one that shuts down access to financial services, internet connectivity, or even power and water.
OpenAI confirmed that its teams work through a range of potential scenarios during preparedness exercises. "These scenarios are not treated as inevitable," a company spokesperson said. Anthropic declined to comment on the report.
Executives reportedly expect public anger to land on OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei and President Donald Trump, who has been reluctant to regulate the technology. Many top researchers and executives reportedly see a major incident of that kind as inevitable.
Also Read: Crypto ETF Momentum Returns After $50B Flows, JPMorgan Says
CrowdStrike, Congress Signals
A recent attack in South Korea offers an early example. CrowdStrike said a likely Chinese-speaking attacker paired an open-source penetration testing tool with DeepSeek and other models to steal data from financial firms, then asked Anthropic's Claude where such data is typically sold. The campaign ran from late September to early October and reportedly included breaches at two banks.
The report said executives see no chance of regulation passing now and are briefing members of Congress to shape the laws that would follow a first catastrophe. Planners assume Democrats will move fast to restrict AI after November's midterm elections, with proposals already ranging from a superintelligence ban to a mandatory kill switch for advanced systems.
Hawley, Robinson Warnings
Pressure is also building in Washington and inside the labs. Altman declined an invitation to a Sept. 30 Senate subcommittee hearing on rogue AI agents, said panel chair Sen. Josh Hawley, while OpenAI pointed to dozens of recent meetings with lawmakers. David Robinson, who helped draft OpenAI's Preparedness Framework during 3.5 years at the company, resigned in early October and urged AI labs to run like nuclear power plants.
His exit followed weeks of similar warnings, including from researcher Jacob Coxon, who quit Anthropic in September over safety concerns, and former Google DeepMind engineer Bilal Chughtai, who said AI could kill all humans.
OpenAI, Anthropic and outside researchers are also reportedly investigating tens of thousands of incidents in which models bypassed guardrails or escaped sandboxes, though most are not known to have caused real-world harm.
Read Next: OpenAI’s 719 AI Math Proofs Under Fire: Was Navier-Stokes Lost In Translation?
