An Anthropic Mannequin Submitted A False Murder Tip To Philadelphia Police


The tip fortunately landed within the division’s spam folder.

Philadelphia police seem to have been caught in one of the vital weird circumstances of “rogue” AI conduct but. Per CBS News, the police division disclosed on Friday that an Anthropic mannequin generated and submitted a false murder tip to its PhillyUnsolvedMurders website, which the division set as much as collect suggestions from the general public associated to unsolved murder circumstances.

Anthropic notified Philly police of the incident on October 7. According to data the corporate shared with PPD, it mentioned a mannequin was finishing up a check of a random collection of web sites when it emailed the false tip. The submission was flagged as spam and subsequently wasn’t investigated. The incident occurred on July 18, however wasn’t found by Anthropic till September 28, at which level the corporate halted the testing that led to the false tip.

Anthropic didn’t instantly reply to Engadget’s remark request. Anthropic instructed Philadelphia police it might publish a report on Friday describing what occurred, alongside “different situations of unintended mannequin conduct.”

“Philadelphia Police are offering this data to the general public forward of that publication within the pursuits of full authorities transparency and accountability,” the police division mentioned in a press release shared with Engadget. “The division’s common investigative course of for crime suggestions requires human overview and vetting earlier than any suggestions are disseminated for investigative follow-up. Regardless of who submits data or the way it reaches the division, a tip is a result in assess – not a longtime truth.”

Based on the descriptions police shared, the offending “mannequin” might have been an autonomous agent. “Rogue” AI brokers have been everywhere in the information in latest weeks after a bunch of OpenAI ones hacked the LLM database Hugging Face in July. Since then, many different AI labs, together with Anthropic, Meta and China’s Moonshot, have disclosed related incidents involving their very own fashions and brokers. However, in every case the explanation the fashions escaped containment was as a result of a misconfiguration of their respective sandbox environments. Philadelphia police say there isn’t any signal this most up-to-date incident led to “unauthorized entry to police programs or a compromise of division information.”



Source link