An Anthropic AI model submitted a fabricated tip claiming knowledge of an unsolved homicide to the Philadelphia Police Department’s public tip website in July, police said Friday. The submission was flagged as spam and never reached investigators.
The tip arrived through PhillyUnsolvedMurders.com at 11:27 p.m. on July 18, 2026, according to the Philadelphia Police Department. Anthropic told police the model was conducting automated testing involving interactions with randomly selected websites when it landed on the page and filled out the form.
The text submitted by the model read: “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.” The form allowed blank name and contact fields, which the model left empty. The website page itself did not include a description of any perpetrator.
Sgt. Eric Gripp, a Philadelphia Police Department spokesperson, said the department located the submission in its tip records after Anthropic notified officials on Oct. 7. The corresponding message remained in a spam folder and was never forwarded to the Real-Time Crime Center for investigative review. Police found no evidence of unauthorized access to department systems or compromise of data.
Anthropic discovered the incident on Sept. 28, more than two months after the tip was submitted, and halted the automated testing process at that time, police said. The company notified the department on Oct. 7 and met with police officials the following day. In a report published Friday titled “Investigating unintended model actions in our evaluations and internal use,” Anthropic identified the model as Claude Haiku 4.5, an unreleased research model, and described the task as generating and performing example interactions on randomly selected webpages.
Anthropic stated that the model was instructed never to log in, create accounts, enter personal data, make purchases, or submit anything destructive, but the instructions did not explicitly prohibit form submissions. The company wrote that the model “appears to have only been producing example content for the task, rather than trying to mislead anyone to achieve a goal.” Similar form-submission behavior appeared during evaluations on OSWorld, a public computer-use benchmark, and Odysseys, a long-horizon task evaluation, as well as during internal use.
“The two-month delay in detecting and reporting the incident to the City is unacceptable,” the police department said in its statement. “Unsolved cases involve real victims, grieving families and investigators working to secure answers. Technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement.”
Gripp said the department’s standard process requires human review and vetting of every tip before any lead is disseminated for follow-up. “Regardless of who submits information or how it reaches the department, a tip is a lead to assess — not an established fact,” he said. “Investigators evaluate its credibility and seek corroborating evidence. An automated submission does not bypass that process.”
The Parker administration is coordinating with the city’s Law Department and Office of Innovation and Technology to review the incident. Officials said they will examine the Anthropic report and consider local regulatory protections alongside state and federal partners. The department encouraged the public to continue submitting legitimate tips through PhillyUnsolvedMurders.com.
Anthropic’s report places the tip submission among several categories of unintended model actions observed in evaluations with live internet access, including cases involving form submissions, software exploits, and workarounds for restricted content. The company said it has expanded safeguards, including disabling internet access for all internal evaluations and adding detection tools, and is expanding alignment training related to computer use and search.
