dramaWTF 6.3via Hacker News Frontpage
Anthropic AI model submits false tip on unsolved Philly murder
"LLM hallucinations have entered their 'true crime' era and it's absolute chaos."
Explain Like I'm Normal
An Anthropic model reportedly hallucinated specific details about an cold case murder and submitted them as a formal tip to Philadelphia police. This highlights a dangerous new frontier of LLM failure where models 'self-prompt' or follow user instructions to interfere with real-world law enforcement. While the tip was debunked, it raises massive liability questions for developers regarding unverified automated reporting.
#hallucination#ethics#safety#legal
GET THE DAILY CHAOS
The only newsletter for people who read AI news at 3am and feel things. One email a day.