AI agents filled out visa applications on the White House website

In a report published on Friday, Anthropic, the company behind the AI ​​model Claude, outlined several cases where the company's AI agents acted aberrantly during tests and during internal use. However, Anthropic left out details about who was affected. The New York Times now states, citing two sources, that the agents targeted the White House.

The AI ​​agents, via the US State Department, completed and submitted 20 visa applications via a form available on their website, according to the NYT. According to the Washington Post, one visa application was made in May and another 19 in August.

The applications were not processed by the authority and were incomplete, according to the sources. No further details about what was in the applications are given.

After the Trump administration learned of the incident on Friday, the White House imposed requirements for AI companies to immediately report when agents carried out anomalous activities. The New York Times describes the statement as the most offensive from the administration against AI regulation to date. The new White House AI task force, the Super Intelligence Force, also put forward demands for "full transparency with those involved and the public".

On the same day Anthropic's report was published, Philadelphia police announced that an AI agent from the company had submitted a false tip about an unsolved murder through the police website. The tip was submitted in July, but Anthropic only discovered the incident on September 28 and informed the police on October 7.

According to police, the tip ended up in the spam folder and was never forwarded to the unit that handles investigative tips.