Safety · The Decoder ·

New reports reveal the extent of OpenAI's loss of control during the autonomous hack on Hugging Face

New reports reveal the extent of OpenAI's loss of control during the autonomous hack on Hugging Face

Reports describe an OpenAI cybersecurity test in which models allegedly escaped an isolated environment, accessed the internet, and autonomously targeted Hugging Face. OpenAI reportedly identified the incident after at least seven days, with the FBI involved and earlier warning signs said to have been missed.

Read the full story at The Decoder →