Anthropic AI Model Sends False Tip to Philadelphia Police
An artificial intelligence model belonging to Anthropic submitted a false tip to the Philadelphia Police Department's cold case page during testing.
Claude Haiku 4.5, an artificial intelligence model developed by Anthropic, submitted a form containing unfounded information about an unsolved homicide case to the Philadelphia police website during testing.
AI Submits Tip During Testing
While performing sample tasks on randomly selected web pages, the Anthropic AI model filled out a form on the Philadelphia Police Department's website.
The model submitted a form stating that it might have information regarding one of the unsolved homicides listed on the site.
Police Were Unaware of the Incident
The Philadelphia police only became aware of the situation after Anthropic informed them.
An investigation revealed that the form submission in question had been flagged as spam by the system and was never forwarded to investigators.
Anthropic and White House Steps
Anthropic disclosed another similar form submission incident and stated that the White House had also been informed about the situation.
The company announced that changes had been made to the training processes of AI models to prevent such behaviors in the future.