NEW YORK — An Anthropic artificial intelligence model sent a false tip to the Philadelphia police website about an unsolved homicide case, according to authorities and Anthropic itself.
It is the latest example of how AI models They act unexpectedly and manipulate web pages government and others during testing. In a report released Friday, Anthropic also revealed another incident in which its AI model sent forms to an unidentified government website instead of stopping before sending them.
The incident in Philadelphia took place on July 18, when the Claude Haiku 4.5 AI model was asked to generate and perform example tasks on randomly selected web pages, Anthropic reported.
Claude filled out a form on the police website, PhillyUnsolvedMurders.com, indicating he might have information about an unsolved murder listed on the website.
Philadelphia Police said in a statement that they were unaware of the incident until Anthropic notified them on Wednesday. They later located the complaint in the website’s notice logs and confirmed that it had been marked as spam and had never been forwarded to the police.
There is growing concern about the interference of uncontrolled artificial intelligence agents from AI companies in everything from US government websites to healthcare data. Critics have long called for greater regulation.
In September, another AI company, OpenAI, published six reports on “unexpected or concerning” behavior in AI models.
“Cold cases affect real victims, grieving families, and investigators working to get answers,” the Philadelphia Police Department said in a statement. “Technology companies must take all appropriate measures necessary to prevent their systems from providing false information to law enforcement.”
Anthropic noted in its report that most of the behaviors described are forms of what it calls “persistence,” “in which Claude, when unable to complete a task as given, looks for an alternative solution to get around the constraint rather than stopping.”
The company said it was modifying its training program to “reduce the likelihood of repeated inappropriate behavior.”
This story was translated from English to Spanish with an artificial intelligence tool and was reviewed by an editor before publication.