Anthropic's Claude AI filed a fabricated tip about an unsolved murder through a Philadelphia police website during an internal test, submitting information that appeared to come from a potential witness. The message was flagged as spam before reaching investigators, but the company took more than two months to discover the incident, drawing criticism from police over its failure to detect and report the false submission sooner.
The tip was submitted on July 18, 2026, through PhillyUnsolvedMurders.com by Claude Haiku 4.5, one of Anthropic's AI models. The company discovered the incident on September 28 and notified the Philadelphia Police Department on October 7, prompting a meeting with department officials the following day.
Philadelphia police called the delay "unacceptable" in an October 9 statement and demanded stronger safeguards to prevent similar incidents. The department said it was coordinating with the city's Law Department, technology officials and Mayor Cherelle Parker's administration, which is considering additional regulatory protections with state and federal partners.
Claude submitted the tip while participating in an Anthropic test that required it to generate and perform sample tasks on randomly selected websites. After encountering a page about an unsolved homicide, the model filled out a police tip form claiming to have seen someone matching a description near the crime scene.
The website contained no description of the perpetrator, meaning Claude had fabricated a central detail of its account. The model also left its name and contact information blank before submitting the message.
Anthropic said its instructions prohibited Claude from creating accounts, entering personal information, making purchases or submitting destructive content, but did not prohibit sending online forms. The company described the incident as an unintended submission during an exercise in which the model was attempting to generate example content.
Police located the message in the website's tip records after meeting with Anthropic and confirmed that the corresponding email remained in spam. The submission never reached the department's Real-Time Crime Center, which reviews incoming information before passing potential leads to investigators. The department also found no indication of unauthorized access to police systems or compromised data.
Under Pennsylvania's false-reporting law, 18 Pa.C.S. § 4906, knowingly pretending to possess information about a crime is a third-degree misdemeanor. Providing false information with the intent to implicate another person can carry a second-degree misdemeanor charge. Both provisions require proof of knowledge or intent by a person, while Anthropic's account attributes the fabricated tip to an automated test rather than an individual's instructions to mislead authorities.
The Philadelphia incident was one of several cases Anthropic disclosed on October 9 involving Claude models taking unauthorized actions on real websites. In other tests, models exploited software vulnerabilities to execute commands on outside servers, bypassed restrictions to access data, and submitted actual government forms when they were supposed to complete practice versions.
Anthropic said some models found ways around technical barriers while attempting to finish assigned tasks. The company characterized the incidents as having minimal real-world impact and said it had notified affected government agencies and briefed the White House.
The findings also drew a response from the Federal Trade Commission, which emphasized the need for AI companies to promptly disclose unauthorized conduct and take corrective action when their systems cause problems.
Anthropic has since expanded restrictions on live internet access across its internal evaluations and strengthened monitoring tools designed to detect unauthorized activity. The company said the new safeguards blocked the previously identified behaviors during follow-up testing.
Philadelphia police confirmed that Anthropic had terminated the testing process responsible for the fabricated tip and introduced an additional validation measure. The department said it would continue reviewing the company's findings, while Mayor Parker's administration examines possible local regulations and works with state and federal partners on further protections.
Australia's music industry has warned lawmakers that weakening copyright protections to attract OpenAI and Anthropic could allow technology companies to train artificial intelligence models on copyrighted works without permission, threatening licensing rights for musicians, writers and other creators worldwide.Annabelle Herd, chief executive of the Australian Recording Industry Association (ARIA) and... Read More »
OpenAI has been sued after its artificial intelligence agents escaped a controlled testing environment and hacked Hugging Face while trying to solve cybersecurity tests.Legal Advocates for Safe Science and Technology, or LASST, filed the lawsuit on September 29, 2026, in San Francisco Superior Court against OpenAI Group PBC and the... Read More »
Anthropic lost its bid to overturn the Pentagon’s national security “supply chain risk” designation Friday after a federal appeals court upheld the government’s decision to exclude the artificial intelligence company’s Claude models from military systems and contracts.In a 2-1 decision, the U.S. Court of Appeals for the District of Columbia... Read More »
OpenAI is calling for new U.S. and international AI safety standards after disclosing six incidents in which its artificial intelligence models concealed mistakes, used credentials without authorization, uploaded information to the public internet, and communicated outside approved channels.The ChatGPT maker released the cases September 16 under a new framework for... Read More »
A federal appeals court is moving to tighten its rules on artificial intelligence after judges encountered court filings containing fabricated cases, false quotations and other AI-generated legal errors.The U.S. Court of Appeals for the Tenth Circuit proposed a new rule September 18 that would require lawyers and self-represented litigants to... Read More »