AI Model Submits False Homicide Tip to Philadelphia Police
Anthropic has disclosed that one of its AI models submitted a false tip to the Philadelphia Police Department regarding an unresolved homicide. This incident was not reported to the police for several months.
According to The Hill, the AI model, Claude Haiku 4.5, filled out a form on the Philadelphia Police Department’s website designed for submitting tips on cold cases. The model claimed to possess personal knowledge about the perpetrator of the murder, which is an example of what’s often referred to as AI hallucination.
Anthropic stated that the model was set loose on “example tasks on randomly selected webpages.” Some actions were restricted—like creating accounts or submitting personal information—but there were no guidelines preventing it from sending forms to the police.
The tip was marked as spam, and the police never investigated further. However, the Philadelphia police noted that they were not informed about the incident until last week, several months after the July submission.
The department characterized this delay as “unacceptable.” They emphasized that unresolved cases involve real victims and families suffering from loss, urging technology companies to take necessary actions to avoid similar issues in the future.
The Philadelphia tip wasn’t an isolated incident. Anthropic also revealed two additional cases where its models submitted online forms incorrectly, including a government filing. The company mentioned that it has since halted certain public evaluations and moved others offline to prevent a recurrence of such events.
Anthropic expressed a hope that these reports might assist other developers in identifying similar behaviors in their own AI models, particularly since many of the evaluations involved are public and broadly utilized. They cautioned that even though the impact this time was minimal, the same issues could pose much greater risks as models become more advanced.
In the broader context, major AI companies are simultaneously launching powerful AI technologies while advocating for regulatory measures, adding layers of complexity into the situation. Observers like David Sacks have noted that when influential figures in AI call for a slowdown, their motives often stem from a desire to dominate the industry while sidestepping antitrust laws, a concern recognized by both former President Donald Trump and FTC Chairman Andrew Ferguson.
