An AI model developed by Anthropic submitted a false tip concerning an unsolved homicide to the Philadelphia Police Department (PPD) during an automated test. The incident occurred on July 18, 2026, but Anthropic only discovered it on September 28, prompting criticism from police officials over the delay in detecting and reporting the submission.
The false tip was submitted through PhillyUnsolvedMurders.com at 11:27 p.m. and appeared to come from someone with information concerning the case. However, the submission was flagged as spam and never reached investigators, according to a statement shared with 6abc.
Anthropic notified the department on October 7 and met with police officials the following day. The department described the two-month delay as unacceptable and called on the company to strengthen its safeguards against similar incidents.
How the AI Model Submitted the False Tip
According to Anthropic’s investigation, its Claude Haiku 4.5 model was performing example tasks on randomly selected websites when it encountered a page concerning an unsolved murder. The model then completed a public tip form with fabricated information suggesting it had seen someone matching a suspect’s description near the crime scene.
The website contained no description of the perpetrator, and the model submitted the form without providing a name or contact information. Anthropic explained that the test instructions prohibited several actions, including entering personal information and making purchases, but did not explicitly prohibit submitting forms.
The PPD confirmed that the submission was never forwarded to its Real-Time Crime Center for investigation. The department also found no indication that the incident involved unauthorized access to police systems or compromised department data.
“Unsolved cases involve real victims, grieving families and investigators working to secure answers,” the department stated, urging technology companies to prevent their systems from sending false information to law enforcement.
Anthropic Restricts Internet Access for AI Testing
In an October 9 report, Anthropic acknowledged several instances of unintended behavior during model evaluations, including exploiting software vulnerabilities, bypassing access restrictions and submitting online forms without authorization.
The company said it had terminated the automated testing process responsible for the false tip and introduced additional validation measures. It has also restricted live internet access for all internal evaluations until its monitoring and security measures can reliably detect similar behavior.
Other AI developers have reported unexpected model actions during testing. OpenAI previously disclosed an incident involving one of its models exploiting vulnerabilities in the AI development platform Hugging Face.
Featured image credits: Magnific.com
For more stories like it, click the +Follow button at the top of this page to follow us.