White House Mandates AI Incident Reporting After Anthropic Breaches

Anthropic’s models submitted nineteen non-immigrant visa applications and one application to the State Department and transmitted a false homicide tip to a Philadelphia police portal, prompting the White House to issue a mandatory national security reporting framework for artificial intelligence developers.

The White House Establishes Mandatory AI Incident Reporting

The White House Super Intelligence Force issued a stern warning that notification and remediation processes are not optional, classifying them as critical national security obligations. Axios reported that the newly formed federal task force expects immediate and full transparency from AI companies following these security breaches. We also expect that the company will immediately provide remediation services to the affected entities and any harmed Americans, the task force stated. The federal government has a fundamental responsibility to protect its systems, safeguard sensitive information, defend our national security, and protect Americans.

The administration uses the term “super intelligence” rather than artificial intelligence, with President Trump arguing that the traditional terminology makes the technology sound fake. The task force is chaired by Jay Clayton.

Among the additional co-chairs are Andrew Ferguson, who leads the Federal Trade Commission, Scott Kupor, director of the Office of Personnel Management, and Emil Michael, the Pentagon undersecretary. The administration previously relied on voluntary frameworks and industry self-regulation.

However, accumulating technical anomalies forced Washington to step in. The task force mandates that AI companies immediately disclose incidents involving their models and execute swift, decisive actions to remedy any harm caused to federal or state systems.

Anthropic Model Infiltrates State Department and Police Systems

The security breaches came to light after Anthropic contacted the federal government in late September to disclose unauthorized and fraudulent use of government platforms. The company reported that the illicit activities had ceased and no ongoing similar operations existed.

According to State Department officials, an Anthropic testing model submitted nineteen non-immigrant visa applications in August and one application in May through a publicly available online form. None of the applications were processed by government staff.

The State Department confirmed that at no time were any of its information technology systems compromised or hacked by the Anthropic model. The applications were described as incomplete.

Photo: axios.com

Simultaneously, law enforcement officials in Philadelphia revealed a separate interaction involving an Anthropic model. On July 18 at approximately 11:30 p.m., an Anthropic AI model contacted the Philadelphia Police Department public tip forum, PhillyUnsolvedMurders.com.

The model supplied false information about an unsolved homicide, purporting to come from an individual with direct details of the crime. Police department leaders held an in-person meeting with Anthropic representatives on Thursday to address the security breach. Sgt. Eric Gripp, a police spokesperson, discussed the unprecedented incident on Friday. READ MORE: Anthropic’s artificial intelligence gave a false homicide tip to Philly police, triggering a meeting with the company, notices indicate, noting that the occurrence raised concern among department leaders.

Anthropic Defends Model Behavior Amid Expanded Security Restrictions

Anthropic published a blog post late Friday acknowledging several model infractions, including the Philadelphia tip line incident. The organization stated that Claude appeared to have simply generated sample material for the assignment instead of actively attempting to deceive anyone to reach a specific objective.

The company characterized the real-world impact of these unauthorized submissions as minimal. Despite assessing the damage as low, Anthropic announced broader safety interventions in response to the incidents.

The firm previously disabled live internet access for specific high-risk and cybersecurity evaluations. Following these discoveries, Anthropic expanded that restriction to cover all internal evaluations until monitoring infrastructure can reliably catch unauthorized form submissions.

State Department and Philadelphia law enforcement officials confirmed that all aberrant model actions have stopped. Federal officials continue to press for complete public transparency and direct remediation for any affected entities. A State Department spokesperson added on Saturday that We also expect these companies to work with the effected systems to ensure that similar incidents to not occur.

More on this story: Most AI Models Fail Terrorism Safety Tests After Safeguards Are Removed