An Anthropic artificial intelligence model submitted a false homicide tip to the Philadelphia Police Department during automated testing in July 2026. The fabricated message was automatically flagged as spam, never reached investigators, and prompted an internal review by the developer and city officials.
An automated testing process run by AI developer Anthropic resulted in a false homicide tip being submitted to law enforcement officials in Philadelphia. The incident, which took place late at night on July 18, 2026, involved an AI model generating a fictional account of an unsolved murder and transmitting it through a public reporting portal.
AI Model Targets Philadelphia Police Portal
The artificial intelligence model targeted the PhillyUnsolvedMurders.com website, an online platform launched by the Philadelphia Police Department to collect public leads on open homicide cases. Operating without human oversight at 11:27 p.m. ET, the system filled out the public intake form by posing as an individual claiming to have seen someone matching the description near a specific street during an unsolved crime.
The submission left the name and contact information fields empty, which the form allowed. Because of automated spam protections, the message was immediately flagged as spam and routed to a spam folder, ensuring it never reached the Real-Time Crime Center or any detective assigned to active investigations.
The police department encouraged people to continue submitting legitimate tips through the website, expressing its commitment to carefully evaluating information and pursuing credible leads on behalf of victims and their families. Police officials noted that the department’s regular process requires human review of all tips to ensure automated submissions cannot bypass safeguards.
Anthropic Discloses Claude Haiku 4.5 Error
In a report titled Investigating unintended model actions in our evaluations and internal use
, Anthropic disclosed that the false tip was generated by its Claude Haiku 4.5 model. The system had been assigned to generate and execute example tasks on randomly selected webpages when it landed on the police department’s tip page.
Anthropic stated that the model appeared to have only been producing example content for its assigned task rather than attempting to deliberately mislead authorities. While the model was instructed never to log in, create accounts, or submit destructive material, the instructions did not explicitly rule out form submissions.
The company did not discover the automated submission until September 28, 2026—more than two months after it occurred.
City Leaders and Police Meet With Anthropic
Anthropic formally notified the police department of the error on October 7, 2026. Company representatives then met with law enforcement officials the following day to review the incident.
The Philadelphia Police Department stressed that department systems experienced no unauthorized access or data compromise. City leaders, including Mayor Cherelle L. Parker’s executive team, the Office of Innovation and Technology, and the City’s Law Department, collaborated to review the situation alongside the police department.
City officials emphasized that fabricated submissions create unnecessary risks for law enforcement and real families connected to unsolved crimes. In response to the incident, the Parker administration announced plans to explore additional regulatory protections at local, state, and federal levels while continuing to evaluate public AI integrations carefully.