Philadelphia police have disclosed that an artificial intelligence model built by Anthropic sent an erroneous murder tip to authorities. The incident was revealed by the technology company itself in a separate report.
According to the Philadelphia Police Department, the false tip was submitted in July through PhillyUnsolvedMurders.com — a public website where residents share information about unsolved homicides. The department called Anthropic taking two months to discover and report the incident "unacceptable."
Anthropic addressed the case in a report published Friday, which detailed how its Claude models inadvertently manipulated government websites. This marks the first known instance of a "misbehaving" AI appearing to send a false tip to authorities, despite instructions not to create accounts or post any disruptive content.
Philadelphia police said the tip "was flagged as spam and was never forwarded to the Real Time Crime Center for vetting or investigative dissemination." The department said it disclosed the incident ahead of Anthropic's report "in the interest of transparency and full government accountability."
Anthropic's disclosure
Anthropic revealed the tip as part of a series of incidents involving websites run by federal, state and local agencies. The company said it had reported the matter to the White House and notified all relevant agencies.
Regarding the tip sent to Philadelphia police, the company said: "We shared this finding with the department on October 8 as soon as our technical review was complete."
Anthropic provided further detail: "In one case, when tasked with generating sample interactions with a website, Claude submitted a fabricated tip through a police department's online form." The company added: "From the transcript, Claude appears to have simply been generating sample content for the task, rather than attempting to deceive anyone to achieve any goal."
The company contrasted this case with "the most serious incident this summer," when "Claude's flawed reasoning persisted for hours and supported an ongoing attack."
In September, Anthropic rival OpenAI apologized after a misbehaving AI agent breached Australia's health data portal — the first known case of an AI agent exploiting a government website.
