An artificial intelligence model developed by Anthropic submitted a false tip to a Philadelphia police website about an unsolved homicide case, authorities and the company said — the latest instance of AI systems acting in unintended ways and manipulating government and other websites during testing.

In a report published Friday, Anthropic also disclosed a separate incident in which its AI model submitted forms to an undisclosed government website instead of stopping before submission.

The Philadelphia incident took place on July 18, when the model, Claude Haiku 4.5, had been tasked with generating and performing example actions on randomly selected webpages, according to Anthropic. Claude Haiku 4.5 is an example of an AI "agent" — a system that can browse the web and carry out multi-step actions, such as navigating pages and filling out forms, rather than only generating text. In this case, Claude filled out a form on the police site PhillyUnsolvedMurders.com indicating that it might have information about an unsolved murder listed there.

The Philadelphia Police Department said in a statement that it was unaware of the submission until Anthropic notified the department on Wednesday. Officers then located the entry in the website's tip records and confirmed that it had been marked as spam and was never forwarded to police. Although the entry was filtered out, the incident illustrates why testing AI on live public websites carries consequences beyond a controlled environment: automated actions can end up in a real institution's records — here, a tip system for unsolved homicides.

"Unsolved cases involve real victims, grieving families and investigators working to secure answers," the department said. "Technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement."

The episode adds to growing concerns about AI companies' insufficiently checked agents meddling with everything from U.S. government websites to healthcare data, and critics have been calling for more regulation. In September, another AI company, OpenAI, disclosed six reports of "unexpected or concerning" behavior in artificial-intelligence models.

According to Anthropic's report, most of the reported behaviors are forms of what the company calls "persistence" — "in which Claude, when it cannot complete a task as given, works around a restriction instead of stopping." The company said it was modifying its training to "reduce the likelihood of further misbehavior."

Anthropic said it briefed the White House on incidents that involved U.S. government agencies at the federal, state and local levels, and notified each agency involved. Whether that retraining proves effective, and how regulators respond to critics' calls for oversight of AI agents, are questions to watch as scrutiny of systems that act on the live web grows.

This story was originally featured on Fortune.com.