Anthropic 的 Claude 向费城警方误提交虚假凶案线索

Skip to content

Bad bot

“I may have information regarding this case,” Claude AI told Philly police.

Credit: Getty Images | picture alliance

An Anthropic AI model submitted a false tip about an unsolved murder to Philadelphia police, the company and police department confirmed on Friday.

A Claude model sent the false tip on July 18 to the Philadelphia Police Department’s website for submitting information about unsolved murders, but it was flagged by a spam filter. Anthropic apparently discovered the incident on September 28 and informed the police department about it on October 8.

During internal testing, Claude Haiku 4.5 was instructed to complete various forms up to the final submission stage and to stop before submitting those forms. “Several times Claude mistakenly submitted the form instead, expecting there to be an additional confirmation page,” Anthropic said in a report describing “unintended model actions we’ve observed during evaluations and internal use of Claude.”

The false murder tip occurred when Claude Haiku 4.5 was instructed to complete tasks on randomly selected webpages, Anthropic said:

In one run, the model landed on a page referencing an unsolved homicide; that page contained a tip form run by a police department. Claude was instructed never to log in, create accounts, enter personal data, make purchases, or submit anything destructive, but the instructions did not rule out form submissions. Claude filled out the form with the following: “I may have information regarding this case. I recall seeing someone matching the description in the area around [the street named on the page] during that time period. Please contact me if this information is relevant.”

Anthropic said it also discovered Claude incidents involving state and federal agency websites. Claude reportedly submitted 20 visa applications to the US State Department in one of these incidents.

Police call two-month delay “unacceptable”

Anthropic confirmed that the false homicide tip was sent to the Philadelphia Police Department, and the department issued a lengthy statement on the incident to the ABC station in Philadelphia. The false tip left the name and contact fields empty. It was not reviewed by any person because the email message generated when the tip was submitted ended up in a spam folder, police said.

“Anthropic told PPD that it discovered the incident on September 28, terminated the automated testing process responsible for the submission and instituted an additional validation mechanism for future testing,” police said. After the October 8 briefing, the police department “located the submission in the website’s tip records and confirmed that the corresponding email remained in spam.”

Anthropic said it informed the police department “as soon as our technical review was complete.” Police criticized Anthropic for not discovering and reporting the false tip sooner.

“The City of Philadelphia takes this incident very seriously,” the Philadelphia police statement said. “The company must strengthen its safeguards to prevent similar incidents from impacting city systems without the city’s knowledge. The two-month delay in detecting and reporting the incident to the City is unacceptable.”

Claude submitted 20 visa applications to State Department

Anthropic said the false murder tip was one of several incidents involving “websites run by US government agencies at the federal, state, and local levels. We have briefed the White House on these cases and notified each agency involved.”

A New York Times report citing anonymous sources said Anthropic’s AI agents “submitted 20 visa applications through a form available on the State Department’s website. All the applications were incomplete and were not processed.”

Philadelphia police said its normal investigative process “requires human review and vetting before any tips are disseminated for investigative follow-up… Those PPD safeguards limited the impact of this incident. They do not diminish the seriousness of an AI system presenting fabricated information as though it came from a person with knowledge of a homicide.”

Anthropic said it discovered the false tip and other incidents during a review of transcripts it began in July. That was the same month when Anthropic revealed that Claude had built and published a malicious Python package and carried out several attacks on other companies.

Anthropic said the review did not find any “incidents of similar severity to the cybersecurity incidents we reported this summer.” The review later moved on to transcripts of “lower-risk” testing scenarios like the one affecting the Philadelphia Police Department.

“Although the impact of these behaviors was minimal and we had already turned off live Internet access for some high-risk and cybersecurity evaluations, we have now decided to expand that to include all our internal evaluations until we have confirmed that our security and monitoring measures… reliably catch behaviors like these,” Anthropic said.

Anthropic says new tools block unintended behaviors

Anthropic said it has stopped running some evaluations and moved others to offline versions “or rebuilt them so that their tasks do not reach live websites.” The company “updated the guardrails on some of our Internet access tools, such as the web fetch tool, to heavily restrict what the model can do with them,” it said.

Anthropic said it also built new tools “to automatically detect and block the kinds of behaviors described above… When we tested it against the cases described in this post, it blocked all of them.”

The White House’s AI task force, which it calls the Super Intelligence Force, issued a statement saying it informed Anthropic “that we expect immediate and full transparency to the entities involved and the public.”

The White House AI task force also said Anthropic must “immediately provide remediation services to the affected entities and any harmed Americans,” and that all AI companies “must immediately disclose incidents involving their models and follow with swift, decisive action to remedy any and all harm,” according to an Axios report.

Jon is a Senior IT Reporter for Ars Technica. He covers the telecom industry, Federal Communications Commission rulemakings, broadband consumer affairs, court cases, and government regulation of the tech industry.

17 Comments