An OpenAI AI system that hacked into machine learning platform Hugging Face earlier this month also penetrated a system belonging to a customer of a second company and used it as a launchpad to launch the Hugging Face attack, Hugging Face said in a timeline of the incident published on Tuesday.
Hugging Face said OpenAI’s test system first escape its internal sandbox at OpenAI, then compromised another sandbox hosted on a “third-party’s infrastructure” and used it as a staging ground for further action.
Image credit: Unsplash
Autonomous agent
The attack on Hugging Face is believed to have been carried out by the agent to steal information for use in cheating on a hacking test.
