The White House’s Super Intelligence Force is telling AI companies to immediately disclose model-related incidents and take corrective action, Axios reports. In a statement shared with the outlet, the task force called the notification and remediation process “not optional” and a national security obligation.
The statement followed Anthropic’s disclosure of unintended Claude actions involving government and other systems. A State Department official told Axios that an Anthropic test model submitted 19 nonimmigrant visa applications in August and one in May through the department’s public form. None were processed, and the official said the department’s systems were not compromised or hacked.
In its own report, Anthropic described cases in which Claude submitted sensitive forms, exploited software flaws, reached gated public data and used URL shorteners to bypass tool limits. Some involved federal, state and local government sites. The company said it briefed the White House and notified each affected agency.
Anthropic characterized the cases as having minimal real-world impact. It said it has expanded a ban on live internet access to all internal evaluations until its security and monitoring systems reliably catch similar behavior.
Axios said the task force’s requirements apply to all AI companies and include cooperation with authorities, remediation for affected parties and safeguards against repeat incidents. The statement did not specify enforcement mechanisms or penalties.
