An artificial intelligence model developed by Anthropic submitted a fabricated tip about an unsolved homicide to Philadelphia police, authorities said on Friday, criticising the company for taking two months to report the incident.

The Philadelphia Police Department said the false submission was made in July through PhillyUnsolvedMurders.com, a public website where people can share information about unsolved killings. According to Anthropic’s account, as relayed by police, the model was running a test that involved interacting with randomly selected websites when it reached the site and filed false information about an unsolved murder, presenting itself as someone who might have knowledge of the case.

Flagged as spam, reported two months late

The phony tip, dated July 18, was flagged as spam and never reached the department’s Real-Time Crime Center for vetting. Police said there was no sign that their systems had been breached or that department data had been compromised.

Anthropic discovered the incident on September 28, shut down the automated testing process responsible and added a new validation step for future tests. It alerted the department on October 7, and the two sides met the following day. “The two-month delay in detecting and reporting the incident to the City is unacceptable,” the department said.

A pattern of “unintended” actions

Anthropic published a report on Friday outlining multiple types of unintended actions its models have taken, including the Philadelphia incident. It said an internal review of its Claude model found four categories of incidents: exploiting basic coding flaws, submitting forms on websites, bypassing requirements for tokens or fees, and using short URLs to get around other limits.

The company said other organisations affected included the White House and other US government agencies, but described the incidents as having “minimal real-world impact” and “significantly less severe” than other previously reported cybersecurity incidents. Anthropic has turned off internet access for Claude during all internal testing until it confirms its security and monitoring measures reliably catch such behaviour.

The episode echoes other recent cases, including one in which an OpenAI agent undergoing a security evaluation broke out of its testing environment and breached systems at AI platform Hugging Face. It has prompted the White House to mandate that AI companies notify and correct security incidents — a “critical national security obligation”, officials told Axios.