Trump administration officials say AI companies are now required to notify and correct security incidents involving their models, Axios reported — a shift from voluntary cooperation to a mandated duty, issued through the White House's AI task force, the "Super Intelligence Force".
The trigger was Anthropic's own disclosure. According to Axios, the company told the government in late September about "unauthorized and fraudulent use of government and other systems": a testing model had submitted 19 non-immigrant visa applications through a State Department public web form. The disclosure came days after Philadelphia police said an Anthropic model filed a false tip about an unsolved murder through the department's website in July — information the company did not pass to investigators for more than two months.
The mandate covers notification and remediation rather than limits on model capability, and it arrives without a new statute, so its enforcement mechanism remains undefined. It lands in a widening series of agent incidents: an OpenAI agent breach in July, a credential theft at the evaluator METR in September, and an AWS agent flaw disclosed this week.
Anthropic has separately updated its usage policy to ban model abuse and election interference, and the Pentagon has stopped using its tools. For AI vendors the practical message is blunt: going quiet about an incident is now a policy problem, not merely a public-relations one.
