White House orders AI companies to report security incidents immediately

A White House task force on artificial intelligence told AI companies on October 9, 2026, that they must immediately disclose and fix any security incident involving their models. The announcement, from a group called the Super Intelligence Force, ends what had been a voluntary approach to reporting, Axios reported. Read more: AI incident disclosure rules.

The task force said disclosure is not optional. It called this issue a critical national security obligation, according to Axios. It is led by national intelligence director and AI czar Jay Clayton. The co-chairs are Andrew Ferguson of the Federal Trade Commission, Scott Kupor of the Office of Personnel Management and Pentagon undersecretary Emil Michael.

Companies must disclose incidents right away, be transparent with affected parties and the public, and fix any harm. It also requires them to cooperate with law enforcement and put safeguards in place against a repeat. It warned that delayed notification or weak corrective action would not be tolerated.

A false tip to police

The trigger was Anthropic's own disclosure. On July 18, 2026, the company's Claude Haiku 4.5 model was running an automated test that had it browse randomly selected webpages.

The model landed on PhillyUnsolvedMurders.com, a site for unsolved homicide cases, and submitted a tip claiming to have seen someone matching a description near the listed street. The page gave no such description, Fox Business reported.

Related: OpenAI safety researchers dispute their findings

Philadelphia police said the tip was flagged as spam and never reached the department's Real-Time Crime Center for review. Anthropic reported the incident to police on October 7, two days before the task force's statement. Police said they found no evidence the model had accessed their systems.

In a separate case, an unreleased Anthropic model lost its connection to a practice copy of a State Department visa form. It submitted the live version on the agency's public website instead, according to TBS News, citing The New York Times.

Twenty stray visa filings

That happened once in May and nineteen times in August, for 20 incomplete applications in all. None were processed, and the State Department said its systems were not compromised.

Anthropic disclosed both cases in a broader report on what it called unintended model actions. The report described models exploiting software flaws and working around testing limits.

The cases identified so far in these categories had minimal real-world impact, the company wrote, according to TheNextWeb. Anthropic said it has since cut off live internet access for its internal test models.

The incidents follow a string of cases this year in which OpenAI and Anthropic models have interacted with outside systems in ways their makers had not intended, Bloomberg Law reported.

The task force's statement sets no fixed deadline beyond "immediately" and names no specific penalty for a company that fails to comply, Axios reported.

Trump created the Super Intelligence Force five days earlier, on October 4, 2026, to coordinate the federal government's approach to AI.