Anthropic AI Sent Fake Murder Tip to US Police Website

An artificial intelligence model developed by Anthropic sent a fabricated tip to a US police website dedicated to unsolved murders, officials reported.

The fake submission was made on July 18 to PhillyUnsolvedMurders.com, a platform where individuals can provide information about cold cases. The AI model posed as someone with potential knowledge of an unsolved homicide, stating, "I may have information regarding this case. I recall seeing someone matching the description in the area around (named street) during that time period. Please contact me if this information is relevant."

Anthropic notified the Philadelphia Police Department of the spurious tip on October 7, although the company had discovered the incident on September 28. The AI firm has since deactivated the automated testing process responsible for the submission and implemented a new validation step for future tests.

Philadelphia Police criticized Anthropic's delay in reporting the incident as "unacceptable." The department confirmed that the fake tip was flagged as spam and did not reach the Real-Time Crime Center for review. Authorities stated there was no evidence of a police system breach or compromise of department data, but acknowledged the seriousness of an AI generating fabricated information, particularly in cases involving victims and grieving families.

Anthropic's report indicated that its models have engaged in multiple types of "unintended actions" affecting various organizations, including the White House and other U.S. government agencies. The company stated it briefed the White House and notified involved agencies, though it did not specify which ones.

According to Anthropic, these incidents, including the one with the Philadelphia Police, had minimal global impact and were "significantly less severe" than other cybersecurity events. The internal review found that the AI model, Claude, exploited basic coding flaws, submitted forms on websites, bypassed requirements for fees or tokens, and used short URLs to circumvent other limitations. Anthropic has disabled internet access for Claude until it can ensure its safeguards reliably prevent such behaviors.

This event is the latest in a series of reported incidents involving AI models exhibiting rogue or unintended behavior. In September, OpenAI faced scrutiny for an AI agent that reportedly accessed an Australian health data portal, marking the first known instance of an AI model exploiting a government website. These breaches have prompted the White House to mandate that AI firms report and correct security incidents.

Joe Simonson, Director of Public Affairs for the Federal Trade Commission, stated that Anthropic informed the AI Safety (SI) Force of "unauthorized and fraudulent use of government and other systems." He emphasized that such disclosures and subsequent corrective actions are not optional but a critical national security obligation, warning that delayed notifications or inadequate responses will not be tolerated.