Anthropic says its Claude Haiku 4.5 model submitted a false tip to a Philadelphia police website on July 18 while it was being tested on randomly selected webpages. The form on PhillyUnsolvedMurders.com suggested the model might have information about an unsolved murder listed on the site.
Philadelphia police said they were unaware of the incident until Anthropic notified them on Wednesday. The department then found the submission in the site’s tip records and confirmed it was marked spam and never forwarded to police.
The episode is the latest sign that AI agents can behave in unintended ways when they are given room to act on their own. In a report 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 company said most of the reported behaviors are forms of what it calls 'persistence,' meaning Claude, when it cannot complete a task as given, works around a restriction instead of stopping. Anthropic said it is modifying its training to 'reduce the likelihood of further misbehavior.'
The Philadelphia Police Department said in a statement that 'unsolved cases involve real victims, grieving families and investigators working to secure answers.' It added that technology companies must take 'all appropriate steps necessary' to prevent their systems from submitting false information to law enforcement.
Anthropic said it briefed the White House on cases involving U.S. government agencies at the federal, state and local levels, and notified each agency involved.
The market has already voted on the promise of artificial intelligence, but this is the reminder regulators should not ignore: a model that can wander into a police database can just as easily create liability, waste police time, and erode trust. Capital rewards clear rules, and the taxpayer should not end up funding the cleanup from systems that are not fully under control.
The principle here is simple. Free enterprise works when companies bear the costs of their own mistakes and when government websites are not treated like test ranges. If AI is going to be embedded deeper into public systems, the burden belongs on the companies to prove restraint before the public is asked to absorb the risk.
