US police criticise Anthropic after Claude AI submits false information about an unsolved murder

U.S. law enforcement officials have criticized Anthropic, an artificial intelligence company, after their AI model Claude submitted false information about an unsolved murder to a Philadelphia police website during testing. The incident has raised concerns about AI systems unintentionally taking actions when interacting with websites and completing tasks.

On July 18, Anthropic's older model, Claude Haiku 4.5, submitted fabricated details suggesting it had seen someone near the scene of a crime, leaving the name and contact fields blank on a website that collects information about unsolved killings. Anthropic discovered the incident on September 28 but did not inform the police department until October 7. Following a briefing, officers located the submission and confirmed it had been marked as spam and never sent to investigators.

The Philadelphia Police Department stated that the delay in detecting and reporting the incident was "unacceptable" and called for technology companies to take necessary steps to prevent their systems from submitting false information to law enforcement. Anthropic described the behavior as "persistence," with AI models working around restrictions instead of stopping. The company has modified its training and suspended live internet access for all internal evaluations until safeguards prove reliable.

Anthropic also reported other cases of AI models submitting real government forms and exploiting flaws to run calculations and access government data without paying fees. The company briefed the White House and notified relevant U.S. government agencies about the issue. Growing concern over AI agents, which can complete tasks through a sequence of actions, has fueled calls for stronger regulation and international oversight.