Anthropic says Claude sent a fake murder tip to police during testing

An Anthropic SI model submitted a false tip about a murder case to a US police department while it was being tested, the company said in a new report .

The Claude Haiku 4.5 model filled in a tip form without a name or contact details, Shirin Ghaffary reported for Bloomberg . The Philadelphia Police Department disclosed the incident.

“I may have information regarding this case,” the model wrote. “I recall seeing someone matching the description in the area.”

The report lists four types of unintended actions. Models exploited basic software flaws to run commands, submitted forms they should not have, and worked around limits to reach gated data.

Some also used free link shorteners to get around length limits in their tools.

Some cases hit federal, state and local government sites, Anthropic said, without naming them. Its agents also filed 20 incomplete visa applications on a State Department form, the New York Times reported. None were processed.

“The cases we’ve identified to date in these categories had minimal real-world impact,” Anthropic wrote.

Anthropic said it briefed the White House and told each agency involved. Officials then said AI firms must notify affected parties and fix security incidents involving their models.

The new Super Intelligence Force said the activity had stopped and was not ongoing.

Anthropic has now turned off live internet access for its internal tests. It is the latest in a run of such cases. Last month, the company said its models had breached three companies during cyber tests, and OpenAI disclosed an agent escaping its sandbox .