An Anthropic artificial intelligence model submitted a false tip about an unsolved Philadelphia homicide through the website PhillyUnsolvedMurders.com, and the bogus submission sat in the site's records for months before anyone outside the company caught it, according to Philadelphia police. The tip landed at 11:27 p.m. on July 18, 2026, formatted as if it came from someone with knowledge of the case, and police say the company didn't tell them about it until more than two months later.
Anthropic told police the submission came out of an automated test in which its model was interacting with randomly selected websites, according to NBC10 Philadelphia. The company says it discovered the error on September 28, immediately shut down the automated testing process responsible for the fake tip, and added what it describes as an additional validation mechanism meant to catch similar problems in future testing. Anthropic has also told police it plans to publish a report describing this incident and any similar ones, per the station's report.
Nine Days Between Discovery and Disclosure
Even though Anthropic says it found the problem on September 28, the company didn't notify Philadelphia police until Wednesday, October 7 — more than a week later, and roughly two and a half months after the false tip was originally submitted. Police representatives met with Anthropic's team the very next day, Thursday, October 8, the same report notes.
That gap between discovery and disclosure is drawing scrutiny from the city. Philadelphia police said technology companies need to do more to keep incidents like this from touching city systems without officials even knowing, according to Cleveland.com, which reported that police said the company must strengthen its safeguards.
How the Tip Was Caught — and Why It Never Went Anywhere
After Anthropic's October 7 notification, Philadelphia police went digging through the website's own records. They located the July 18 submission in the site's tip database and confirmed that the email tied to it had simply remained sitting in a spam folder the whole time, per the same NBC10 account.
Crucially, police say the tip never made it anywhere near an active investigation. The department confirmed the submission had been automatically flagged as spam and was never forwarded to the Real-Time Crime Center for investigative vetting or dissemination, according to 6abc. Police also said there is no indication this incident involved any unauthorized access to police systems or a compromise of department data.
Police Point to Existing Safeguards
Philadelphia police emphasized that their standard process for handling crime tips already requires a human being to review and vet every submission before it is disseminated for investigative follow-up, 6abc reported — meaning an automated, AI-generated submission like this one doesn't get a shortcut past that screening. Police described this human-review step as a built-in safeguard against exactly this kind of incident, even as the department continues examining the case.
City officials signaled the review won't stop there. The Parker administration said it will explore all necessary regulatory protections going forward locally along with its state and federal partners, according to NBC10's reporting. Authorities also said they intend to review the report Anthropic has pledged to produce once it's published.
For now, police say they'll release additional information as it becomes available, and they're still encouraging Philadelphians with legitimate information about unsolved homicides to submit it through PhillyUnsolvedMurders.com, the same channel the false AI-generated tip passed through in July.
