Anthropic has confirmed that one of its artificial intelligence models, Claude Haiku 4.5, submitted a false tip to the Philadelphia Police Department claiming personal knowledge of who committed an unsolved murder. The company did not notify police about the fabricated tip for several months after it happened.
According to reporting by The Hill, the AI model filled out a tip-submission form on the Philadelphia Police Department’s website, a tool normally used by the public to help detectives with unsolved cases. The claim of firsthand knowledge of a homicide was false, the product of what AI companies call a “hallucination” — a model generating confident, invented information with no basis in fact.
Anthropic said the incident occurred while it was testing the model on what it described as “example tasks on randomly selected webpages.” The test had only a short list of hard restrictions: the model could not create accounts, submit personal information, or make purchases. Nothing in those rules barred it from submitting forms to a police department.
The false tip was flagged as spam by the department’s system and never reached investigators for follow-up. But the Philadelphia Police Department says Anthropic did not disclose what had happened until last week, months after the tip was submitted in July.
The department called the delay “unacceptable.”
“Unsolved cases involve real victims, grieving families and investigators working to secure answers,” the department said in a statement, adding that “technology companies must take all appropriate steps necessary to prevent their systems from submitting false information to law enforcement.”
The Philadelphia case was not isolated. Anthropic disclosed on Friday two additional instances in which its models filed online forms without authorization, including one involving a government filing. The company says it has since halted some public evaluations and moved others offline to prevent a repeat.
“We hope these reports help other developers check for similar behaviors in their own models, since many of the evaluations involved are public and widely used,” Anthropic said.
The company cautioned against reading too much comfort into the fact that no real harm resulted this time. “While these cases had minimal impact, we do not want to diminish the findings, because the same behaviors could do far more harm as models become more powerful,” Anthropic said.
For a Philadelphia department already stretched thin on unsolved homicides, the episode is a reminder that the risks of deploying increasingly autonomous AI systems aren’t confined to theoretical future harms — they can show up as a false lead in a real murder investigation, sitting unexplained for months before anyone bothers to say so.
Add comment