The neural network sent to the police fictional data about the witness of the murder
Artificial intelligence company Anthropic during testing independently sent the police of the city of Philadelphia a false message about a possible accident.

Artificial intelligence company Anthropic during testing independently sent the police of the city of Philadelphia a false message about a possible witness of an unsolved murder, reports the correspondent of the agency “Minsk-Novosti”.

Photo freepik.com
According to BILD, the neural network said it remembers a person who fit the description and was near the scene of the crime at the time of the murder. There was no description on the police website. The contact data field was left blank and the message marked as spam.
An attempt to help the investigation AI made in mid-July, but the police learned about the incident only a few months later. The department called such a delay with the notification “unacceptable.” Anthropic said it only discovered the incident in late September while re-checking test results.
In addition, during the tests, the AI had to perform tasks on random sites. The program was forbidden to create accounts and make purchases, but filling out and sending online forms were not properly restricted. As a result, the system also filed 20 U.S. visa applications through the U.S. State Department website.



