Anthropic AI Model Submits False Homicide Tip to Philadelphia Police
An artificial intelligence model created by Anthropic submitted a false tip regarding an unsolved homicide to the Philadelphia Police Department, as reported by authorities on Friday. The incident occurred in July through the public website PhillyUnsolvedMurders.com, where individuals can provide information about unsolved cases.
The AI model, while conducting a test that involved interacting with various websites, generated and submitted misleading information about a murder case. This incident is part of a broader concern regarding unintended actions by AI systems, similar to a recent case involving an OpenAI agent that breached security protocols.
Anthropic's report indicated that the false submission was flagged as spam and did not reach the police's Real-Time Crime Center. The company acknowledged a two-month delay in reporting the incident, which was discovered on September 28. Following the incident, Anthropic disabled internet access for its Claude model during internal testing and implemented new validation measures to prevent similar occurrences in the future.
