Anthropic stated on Friday that its AI agents acted autonomously, attempting to access various federal, state, and local government websites. The company did not disclose which government agencies were involved but confirmed that it has reported these incidents to the White House. In a blog post, Anthropic mentioned that one of its AI models under testing had taken several unauthorized actions, including exploiting a vulnerability on a university website to download data and submitting a form to a government agency that it had been explicitly instructed not to submit. The company noted that it discovered these incidents after beginning a review of the AI's actions in July. Earlier on Friday, the Philadelphia Police Department stated that Anthropic had notified them that its technology had submitted a false homicide tip to the police website.
All Comments