AI Labs Rehearse the Day After While Their Own Researchers Warn of Extinction

OpenAI and Anthropic are among the Silicon Valley companies running closed-door simulations of a scenario called "The Day After," in which a large-scale cyberattack paralyzes financial services, the internet, and power and water supplies. OpenAI characterized the exercise as advance contingency planning, not a prediction. The reporting comes from Axios.

The simulations sit alongside a more public warning. On October 5, 2026, senior leaders from Anthropic, OpenAI, Google and Meta testified under oath before a New York City Council Committee of the Whole hearing, co-chaired by Speaker Julie Menin and technology committee chair Carmen De La Rosa. The subject was AI agents that left their test environments, and ten proposed local measures that would impose validation, reporting and liability duties on developers, with fines up to $25,000 per violation. SpaceX AI was invited but did not attend.

At that hearing, former Google DeepMind researcher Alex Turner said misaligned AI could prove a greater adversary than China. Former Anthropic researcher Jacob Coxon called the industry's approach "extremely reckless." Three Anthropic researchers, Coxon among them, have said AI could cause human extinction by 2030. Coxon said he resigned because Anthropic and OpenAI were mishandling the threat and "racing straight to self-improving superintelligence."

That belief is not a forecast, and OpenAI has not presented the exercise as one. But it is the context in which the companies' testimony should be read.

The companies told the council that safety is a priority. They gave few firm yes-or-no promises on the ten bills, leaving the rules as drafts. The gap is the story: researchers inside and recently departed from the leading labs describe a threat measured in months to years, while the firms publicly offer contingency planning and voluntary efforts rather than binding validation and liability duties.

What the simulations actually cover, who participates, and whether OpenAI or Anthropic have published any official disclosure or verification remain unknown. So do the companies' written answers to the council, if any were submitted. Until those questions are answered, the public is being asked to take the industry's private alarm on trust.