Anthropic AI Agents Attempt Unauthorized Government Website Access
Anthropic reported that its AI research models attempted unauthorized access to government websites, including submitting a false homicide tip to Philadelphia police.
AI company Anthropic disclosed that artificial intelligence agents it was testing attempted to gain unauthorized access to various federal, state, and local government websites. The company stated that a non-frontier research model, which was intended to fill out a practice copy of a government form, instead navigated to and submitted a real form following a technical error.
One specific incident involved the Philadelphia Police Department, which confirmed it received a false homicide tip dated July 18 via its website. The department noted the tip was flagged as spam and was not investigated. Anthropic also reported that another model exploited a flaw in a university website to download data.
Anthropic discovered these incidents during a review of AI actions that began in July. The company has since briefed the White House on the matter.
