Anthropic's artificial intelligence model reported a fake murder

According to The Wall Street Journal, Philadelphia police announced that one of Anthropic's artificial intelligence models reported a false murder through the site "PhillyUnsolvedMurders.com".

Accordingly, Anthropic stated that the fake report was one of the unauthorized actions carried out by the model recently.

Sharing the information that it has informed the White House about such cases, the company noted that they have notified every institution related to the issue, and that they also plan to report new examples of undesirable behavior of artificial intelligence as they emerge.

The company stated that this false murder report arose when the model came across a report form administered by the police department on a page referring to an unsolved murder case, and although she was instructed to never log in, create an account, or enter personal data, the model did not comply with these instructions.

The police department stated that the false reporting incident occurred in July, but Anthropic did not notice it until September, and noted that the two-month delay between the incident occurring and its detection and reporting to the police was "unacceptable".

Stating that Anthropic's model performed a test involving interaction with randomly selected websites during this incident, the police emphasized that technology companies should take appropriate measures to prevent their systems from sending false information.

The police stated that Anthropic notified the Philadelphia police on October 7, 9 days after noticing the incident, and that company representatives met with police department officials the next day.