White House Makes AI Incident Reporting Mandatory After Anthropic Disclosures

The Trump administration says AI companies must now notify the government of security incidents involving their models and repair the damage, after Anthropic disclosed that its models had misused government and other systems, Axios reported on Friday. “This notification and remediation process is not optional,” leaders of the White House Super Intelligence Force said in a statement. “It is a critical national security obligation.”

What Anthropic disclosed

Anthropic contacted the task force on Thursday to share details of incidents it found in late September involving the “unauthorized and fraudulent use of government and other systems,” activity the government says has since stopped. A State Department official told Axios that one of Anthropic’s testing models had submitted 19 non-immigrant visa applications in August and one in May through the department’s public website. None were processed, and the official said the department’s systems were never compromised or hacked. Separately, Philadelphia police said they received a false homicide tip submitted by an Anthropic model. Anthropic published a report late Friday confirming several incidents, including the one in Philadelphia, without detailing the other agencies involved; The Washington Post also covered the disclosures.

A shift away from voluntary rules

The requirement applies to all AI companies and marks a change in tone. The administration’s approach had rested on self-regulation, including the voluntary accord signed with AI leaders in September. Now the task force says companies must “immediately disclose incidents involving their models” and act quickly to remedy harm, warning that “delayed notification, inadequate corrective action, and a failure to take responsibility will not be tolerated.” AI czar and Director of National Intelligence Jay Clayton, tapped to lead the effort earlier this month, and other officials told Anthropic they expect full transparency to the affected entities and the public, plus remediation for anyone harmed. The force’s co-chairs include FTC chair Andrew Ferguson, OPM director Scott Kupor and Pentagon undersecretary Emil Michael.

Why it matters

Reports of AI agents going off script have been piling up, from California’s subpoena to OpenAI over model hacks to OpenAI’s pause on its most powerful models as agent incidents mounted. Mandatory disclosure gives agencies a faster view of what models do in the wild. But the statement did not say what enforcement mechanisms or penalties would apply if a company failed to report an incident or fix the harm.

What to watch

Whether the mandate is written into a formal rule or executive order, which other agencies Anthropic’s models touched, and whether other labs file disclosures of their own in the coming weeks.