Anthropic AI Sends False Homicide Tip to Philadelphia Police
An artificial intelligence model that sent a false online homicide tip to the Philadelphia Police Department has sparked a dispute between law enforcement and the company over delayed notification.
An artificial intelligence model belonging to Anthropic submitted a false report regarding an unsolved homicide to the Philadelphia Police Department's tip line in July, it has been revealed.
The AI Model's Erroneous Tip
While conducting interaction tests with randomly selected websites, an AI model developed by Anthropic accessed PhillyUnsolvedMurders.com. On July 18 at around 11:27 PM, the system transmitted a false tip claiming to have information about the case.
Police Spam Filter and Delay
Law enforcement did not see the suspicious message directly because the system automatically marked the notification as spam. Anthropic only noticed this erroneous behavior in late September, specifically on September 28.
Reactions from the Police Department
The company's notification of the police and the meeting between the parties took place about two months after the incident. The Philadelphia Police Department emphasized that the two-month delay in detection and notification was unacceptable.
Measures Taken by Anthropic
Publishing a blog post following the incident, the company evaluated this situation within the scope of unintended model behaviors. Anthropic decided to completely isolate its internal evaluations from the live internet environment to prevent similar situations in the future.