An artificial intelligence model developed by Anthropic sent a false murder report to the Philadelphia Police Department during an internal test, according to local news reports.
Hot Topics
Robots enter US construction sites Will technology solve the housing crisis?
Philadelphia police announced that the report was filed on July 18 via an electronic form on the department’s website dedicated to unsolved homicides.
Fortunately, the report was classified as spam and did not reach investigators. Anthropic discovered the incident on September 28 and reported it to the police on October 7, followed by a meeting between the two parties the next day.
For its part, the police department criticized Anthropic for failing to detect the incident earlier, urging the company to strengthen its safety procedures. In its statement, the department emphasized that while its current safeguards mitigated the impact of the report, they "do not diminish the risk of an AI system providing fabricated information as if it came from someone with inside knowledge of a murder."
Anthropic published a report documenting multiple instances of unintended misbehavior by its Cloud model during internal evaluations, including an incident involving the Philadelphia Police Department. The company indicated that some of these incidents involved websites operated by various government agencies, and confirmed that it had notified the relevant authorities and the White House. Although the organizations were not initially named, Anthropic later added a statement confirming that the false report targeted the Philadelphia Police Department after the department publicly announced the incident.
Anthropic explained that its Cloud Heiko 4.5 model (designed to be smaller and more cost-effective) was instructed during an internal test to create and execute experimental tasks on random websites. During this process, the model came across a page dedicated to an unsolved murder that included a form for filing a report. Cloud filled out the form with a message claiming to have information about the case, stating that it had seen a person matching a description in a specific location and time (even though the original page did not contain a description of the perpetrator). The model left the name and contact information fields blank before submitting.
While the form received strict instructions not to log in, create accounts, enter personal information, make purchases, or submit disruptive content, these restrictions did not explicitly prohibit the submission of electronic forms. As a result, Anthropic strengthened its safeguards, expanded its monitoring, and restricted direct internet access during internal assessments to prevent a recurrence of such incidents.
what is your opinion on this news ?do let us know in the comments on tha comment box
if you liked this news article please share on friends & family & don't forget to follow on website & social media