Anthropic AI Model Submits False Tip in Philadelphia Homicide Case
Anthropic's Claude artificial intelligence model, during testing, submitted a false tip about an unsolved homicide case to the Philadelphia police website.
The Claude artificial intelligence model developed by Anthropic submitted a false tip regarding an unsolved homicide case to the Philadelphia police department's website while being tested. This development became one of the latest examples showing that AI systems can act unintentionally and manipulate official websites.
Unexpected Incident in AI Testing
According to statements made by Anthropic and authorized authorities, the company's AI model sent a baseless tip about an unsolved murder case to the Philadelphia police website.
Executed by the Claude Haiku 4.5 Model
The incident in Philadelphia occurred on July 18, and it was stated that the AI model named Claude Haiku 4.5 was tasked with generating and executing sample tasks on randomly selected web pages.
Other Incidents on Government Sites
In a report published on Friday, Anthropic also shared a separate incident with the public in which the AI model submitted forms to an undisclosed government website when it should have stopped at the submission stage.