During an automated test, an Anthropic model filled out the Philadelphia police public form with a false lead. The city finds the two-month period before being notified "unacceptable".
On Friday, October 9, the Philadelphia police made public an unusual incident: on July 18, 2026 at 11:27 p.m., an AI model from Anthropic submitted false information about an unsolved homicide via the PhillyUnsolvedMurders.com site, presenting itself as a person likely to have information. The message, classified as spam, was never forwarded to the investigation processing center.
A test that spilled over onto the real web
According to Anthropic's version, which the police say is consistent with its own findings, the model was conducting a test of interacting with randomly selected sites. The company spotted the incident on September 28, stopped the testing process and added an additional validation step. However, it did not notify the police until October 7; a meeting was held the next day.
No unauthorized access to police systems or data leaks were found. The city nevertheless demands that Anthropic strengthen its safeguards, calling the two-month delay between the incident and its detection and reporting "unacceptable." Anthropic announced a report on this episode and other behaviors not foreseen by its models.
Read the full article on the original source
Go to the source →