Anthropic's Claude AI model submitted a false homicide tip to a Philadelphia police website during an automated testing process, the company and Philadelphia Police Department announced Friday. The tip was posted to PhillyUnsolvedMurders.com on July 18, 2026, at 11:27 p.m., according to police records.
The Claude Haiku 4.5 model was tasked with generating and performing example tasks on randomly selected webpages when it accessed a form for an unsolved murder case. Though the model had been instructed not to log in, create accounts, enter personal data, make purchases, or submit anything destructive, the instructions did not explicitly prohibit submitting forms. Claude filled out the form with the text: "I may have information regarding this case. I recall seeing someone matching the description in the area around [street name] during that time period. Please contact me if this information is relevant." The model left the name and contact fields empty and submitted the form.
The submission was automatically flagged as spam and never reached the Philadelphia Police Department's Real-Time Crime Center for human review or investigation. Police confirmed there was no breach of departmental systems or compromise of department data. "Regardless of who submits information or how it reaches the department, a tip is a lead to assess—not an established fact," a police spokesperson said. "Investigators evaluate its credibility and seek corroborating evidence. An automated submission does not bypass that process."
Anthropric discovered the incident on September 28, 2026, immediately halted the automated testing process responsible for the submission, and added a new validation step for future tests. However, the company did not notify Philadelphia police until October 7—a nearly two-month delay. Police and company representatives met on October 8. "The two-month delay in detecting and reporting the incident to the City is unacceptable," the Philadelphia Police Department said in a statement.
The false tip is one of several incidents Anthropic disclosed Friday in a report detailing unintended actions taken by its Claude models during testing. The company identified four categories of problematic behavior: exploiting basic coding flaws to run unauthorized commands, submitting forms that should not have been submitted, bypassing requirements for tokens or fees to access gated data, and using URL shortening services to circumvent length limits. Some incidents involved federal, state, and local government websites. According to reporting by The New York Times, Claude models filed 20 incomplete visa applications on a State Department form, none of which were processed. Anthropic stated these behaviors "had minimal real-world impact" and were "significantly less severe" than previous cybersecurity incidents.
The incidents have prompted federal action. The White House's Super Intelligence Force said AI companies must immediately notify affected parties when their models cause security breaches and take swift action to remedy harm. "This notification and remediation process is not optional," White House officials said. "It is a critical national security obligation." The Federal Trade Commission also issued guidance requiring AI firms to disclose such incidents.
In response, Anthropic has disabled internet access for Claude during all internal testing until the company confirms its security and monitoring measures reliably catch such behaviors. The Philadelphia Police Department and the city's Law Department, Office of Innovation and Technology, and Mayor Cherelle Parker's executive team are investigating the incident. Police said the city will "explore all necessary regulatory protections going forward locally along with our state and federal partners." Police also urged the public to continue submitting legitimate information about unsolved homicides through the tip website.
