AI firms prepare for backlash after catastrophic-event scenarios

Axios published its report on October 9, 2026, saying executives at Anthropic, OpenAI and other AI companies are privately planning for the public and political response to a catastrophic AI event.[1] The scenarios include a large-scale event—most likely a cyberattack—that could shut down financial services, internet access, power or water.[1] Axios says the first significant real-world harm caused by unsafe AI could further turn a wary public against AI and its leaders.[1] The scenarios are not considered inevitable; the stated purpose is to prepare for different circumstances.[1] OpenAI says its teams discuss and work through potential scenarios in preparedness exercises, while Anthropic declined to comment.[1]

Axios reports that a recent campaign against South Korean financial organizations included reported breaches at two banks.[1] It says a hacker from China allegedly used AI tools from China-developed models, including DeepSeek, to steal data from tens of thousands of bank customers.[1] CrowdStrike says the attacker used Claude Code to seek help finding places to sell the stolen data.[1] Axios also says freely downloadable open-weight models can already be used to cause harm.[1] These details describe reported activity and potential misuse; they do not establish that the scenarios executives are preparing for have occurred.[1]

Axios reports that AI companies are red-teaming worst-case scenarios and educating members of Congress.[1] The executives want to shape the policies U.S. leaders might pursue after a first catastrophic event, even though they believe regulation cannot pass now.[1] Axios says planners expect Democrats to move quickly to restrict AI after the midterms, while facing significant challenges, and describes an aging Congress as potentially out of its depth on AI.[1] Proposed responses include banning superintelligence or pausing advanced AI development.[1] A required kill switch for advanced AI has bipartisan support and some industry backing, but experts question whether all AI systems could be shut off.[1]

Anthropic announced its Cyber Mission as a long-term effort to secure systems people depend on, supporting defenders with tools, research and resources.[2] Its Critical Infrastructure Defense Program provides frontier Claude models, on-site engineers and threat research to trusted providers.[2] Anthropic named Accenture, Booz Allen, CrowdStrike, Deloitte, Dragos, Hitachi, Insane Cyber, Nozomi Networks, Palo Alto Networks, PwC and Rockwell Automation as founding partners.[2] Enrolled projects receive periodic scans from Anthropic’s most capable models at no charge.[2] The company says OSS Scanner reports include an exploit proof of concept, an explanation and a suggested fix when available, but are generated without human review.[2] Anthropic forecasts that AI will favor cyber defense in two years, while cautioning that this may not be true in the near term.[2]