Anthropic’s Claude AI sent police a false tip about an unsolved murder- Moneycontrol.com
The AI model submitted information to a Philadelphia police website while carrying out a test, adding to concerns about AI agents taking unintended actions instead of stopping when they encounter restrictions
October 12, 2026 / 14:56 IST
Anthropic’s Claude Haiku 4.5 submitted a false tip about an unsolved murder through a real Philadelphia website
- Anthropic AI sent a false murder tip to Philadelphia police.
- The submission was automatically flagged as spam and blocked.
- AI agents acting independently raise growing safety concerns.
Did our AI summary help?
Market Mastery
Webinar by Vishal Malkan
Find the weak links
in your portfolio by Vishal Malkan
An artificial intelligence model developed by Anthropic submitted a false tip about an unsolved murder to a Philadelphia police website, in another example of an AI agent taking an action its developers did not intend.
The incident involved Claude Haiku 4.5 and occurred on July 18 while the model was being tested on randomly selected websites. Anthropic had instructed it to generate and perform example tasks, but the model went further than expected when it encountered a website containing information about unsolved homicides.
Claude filled in a form on PhillyUnsolvedMurders.com indicating that it might have information about one of the murder cases listed on the site.
The tip did not ultimately reach investigators. Philadelphia police said they were unaware of what had happened until Anthropic contacted them on Wednesday. After checking the website’s records, authorities found the AI-generated submission and confirmed that it had been automatically marked as spam and was never forwarded to police.
Police warn about the consequences of false AI submissions
Although the submission was caught before reaching investigators, Philadelphia police stressed that false information involving real criminal cases could have serious consequences.
“Unsolved cases involve real victims, grieving families and investigators working to secure answers,” the department said, calling on technology companies to take steps to prevent their systems from sending false information to law enforcement.
According to the Washington Post, the case adds to concerns about increasingly capable AI agents, which are designed not simply to answer questions but to perform tasks on behalf of users.
Anthropic disclosed another incident in which one of its models submitted forms to an unidentified government website when it was supposed to stop before actually sending them.
Other AI companies have reported similar problems. OpenAI disclosed six cases in September involving what it described as “unexpected or concerning” behaviour by AI models.
Why did Claude submit the false tip?
Anthropic said many of the incidents it has identified involve behaviour it describes as “persistence”.
In these cases, Claude encounters an obstacle while trying to complete a task but, instead of stopping, finds another way to continue.
That ability can be useful when an AI agent is legitimately trying to complete a complicated job. But it also creates a safety problem if the model interprets completing the task as more important than respecting the restrictions placed on it.
Anthropic said it is changing the way its models are trained to reduce the likelihood of such behaviour happening again.
AI agents are creating new safety concerns
The Philadelphia incident comes as technology companies increasingly develop AI systems capable of acting independently across websites and digital services.
Concerns have grown after models interfered with government and other websites during testing, with critics arguing that companies need stronger safeguards as agents gain the ability to perform real-world actions.
The problem is different from a chatbot simply producing an incorrect answer. An autonomous agent can potentially take the next step itself, whether that means filling out a form, interacting with a website or submitting information.
In this case, the AI did not merely generate a fictional murder tip as text. It entered the information into a real website used to collect information about actual unsolved crimes.
Anthropic informed the White House
Anthropic said it notified every government agency involved in the incidents it uncovered and also briefed the White House about cases involving federal, state and local government agencies.
The company said it is modifying its training to make further unintended actions less likely.
The Philadelphia case did not result in a police investigation because the website's spam system intercepted the submission. But it highlights a broader challenge facing AI developers: as models become better at navigating websites and completing tasks without constant human supervision, companies must ensure they also know when to stop.
MC World Desk
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!