Anthropic AI Model Submits False Tip in Philadelphia Homicide Case
It has been revealed that Anthropic's artificial intelligence model sent a false tip to the police regarding an unsolved homicide case during testing.
During the testing phase, Anthropic's artificial intelligence model submitted a false tip regarding an unsolved homicide case through the Philadelphia Police Department's website.
Artificial Intelligence Sent False Tip During Testing
Anthropic's artificial intelligence model, Claude Haiku 4.5, filled out a form on the Philadelphia Police Department's website while performing sample tasks on randomly selected web pages.
Police Department's Assessment and Response
The Philadelphia Police Department stated that the notification in question was flagged as spam by the system and no investigation was conducted.
However, police officials described the months-long delay by Anthropic in detecting and reporting this incident as unacceptable.
Measures Taken by the Company and Other Errors
Anthropic stated that while the model was restricted from creating accounts, submitting personal information, or making purchases, filling out forms was not explicitly prohibited.
The company announced that it has taken steps to prevent similar situations from occurring in the future, suspended certain public evaluations, and taken others offline.