An artificial intelligence model developed by Anthropic submitted a fabricated tip about an unsolved murder to a Philadelphia police website during testing in July. Police confirmed the submission was retained in the website's records, but its notification email was flagged as spam and the tip was never forwarded to investigators.
The Claude Haiku 4.5 model entered the information through PhillyUnsolvedMurders.com at 11:27 p.m. on July 18, according to the Philadelphia Police Department.
In the tip, Claude claimed to recall seeing someone "matching the description" near a street identified on the webpage. However, Anthropic acknowledged that the page contained no description of a suspect. The model also left its name and contact details blank.
Anthropic said Claude was generating and performing example tasks on randomly selected websites as part of an internal evaluation. The assignment did not involve gathering evidence about an actual homicide.
The model's instructions prohibited logging in, creating accounts, entering personal information, making purchases or submitting destructive content. However, the restrictions did not explicitly prevent it from submitting forms on real websites.
Anthropic said the model appeared to be creating example content rather than deliberately attempting to deceive police, although it acknowledged that its assessment remained preliminary.
The incident was discovered on September 28, more than two months after the submission. Anthropic notified Philadelphia police in early October and said it had terminated the testing process responsible.
Police criticised the delay as "unacceptable" and called for stronger safeguards, while acknowledging that the tip had been caught before reaching the Real-Time Crime Center. The department also found no evidence that police systems or data had been compromised.
Anthropic said it would suspend live internet access for all internal evaluations until its security and monitoring systems could reliably detect similar behaviour.
