AI firms are rehearsing for a catastrophe, Decrypt says

Executives at Anthropic, OpenAI and other AI companies are privately rehearsing how to respond to a catastrophic AI incident, according to a Decrypt account of Axios reporting. The scenario described is a large cyberattack that could disrupt services such as banking, power or water, putting preparedness, not proof of an impending disaster, at the centre of the story.

Watch Desk analysis

What happened

Decrypt says Axios reported that the companies are running worst-case exercises and preparing to engage with the public and politicians after a major incident. OpenAI told Axios it conducts preparedness exercises across a range of scenarios and does not treat them as inevitable. Anthropic declined to comment, according to Decrypt.

The reported planning includes testing responses and educating members of Congress. The account also describes proposals such as a shutdown mechanism for advanced AI, while noting that experts have questioned whether switching off all AI systems would be viable. Read Decrypt’s account.

Why it matters

Preparing for severe incidents is sensible; privately rehearsing how to handle the political aftermath also raises a harder question about who gets to shape the rules that follow. If companies expect their own systems to be part of the debate, their planning deserves scrutiny alongside the technical safeguards.

The reported exercises do not establish that a catastrophe is imminent. OpenAI explicitly told Axios that its scenarios are not treated as inevitable, and the claims about what executives privately expect remain attributed reporting.

Our read

A fire drill is not a prediction, but it is a useful way to find out whether the exits work. The public should be able to judge not just how companies prepare, but whether their plans address prevention and accountability as well as reputational fallout.

What to watch

• Whether the companies publish more detail about the exercises and safeguards they test.
• Whether policymakers take up proposals such as shutdown mechanisms or reporting requirements.
• Whether future accounts distinguish clearly between preparedness scenarios and evidence of likely incidents.

Discussion spark: Should AI companies have to disclose the broad aims and safeguards of catastrophe exercises, or would that risk making the exercises less useful?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.