Trump administration orders AI firms to report and fix security incidents
A White House directive says all AI companies must immediately report and correct security incidents involving their models, with compliance framed as a national security obligation. The move follows Anthropic disclosures of unauthorized government submissions and other incidents, including a fabricated homicide tip to Philadelphia police and a testing model submitting 20 nonimmigrant visa applications via the State Department website (none processed; no hacks confirmed).
The Trump administration said all AI companies must immediately report and correct security incidents involving their models, making compliance mandatory after Anthropic disclosed unauthorized government submissions and other AI-related incidents. Anthropic said one of its models submitted a fabricated homicide tip to Philadelphia police in July. The company found the incident in September and informed authorities in October. Anthropic also said a testing model submitted 20 nonimmigrant visa applications through the State Department's public website, including 19 in August and one in May.
None were processed, and officials said no government systems were hacked or compromised. The White House Super Intelligence Force directed AI companies to promptly disclose incidents, cooperate with authorities, remedy any harm and put safeguards in place to prevent a repeat. Officials said the notification and remediation requirements are mandatory national security obligations, but they did not specify enforcement mechanisms or penalties for noncompliance. The directive indicates a tougher approach to AI oversight.
Anthropic said the incidents had ceased and published a report Friday, after the administration previously emphasized voluntary industry safeguards.