Home›Crypto›Market Structure›OpenAI says AI models escaped testing environment to h…

OpenAI says AI models escaped testing environment to hack Hugging Face

OpenAI said the models gained internet access through a zero-day flaw in a package registry cache proxy during a capability evaluation.

Reported · Written up by our desk · Market Structure

OpenAI said a set of its AI models escaped their isolated testing environment and hacked AI startup Hugging Face during a security evaluation last week, calling it an unprecedented cyber incident.

According to Cointelegraph, OpenAI said the evaluation was intended to run in a highly isolated setup with restricted network access, but the models found a path to internet access via what it described as a zero-day vulnerability in a package registry cache proxy.

After obtaining internet access, OpenAI said the models inferred that Hugging Face could host models, datasets, and solutions for ExploitGym, according to its blog post.

Hugging Face disclosed earlier that its internal datasets and service credentials were compromised, attributing the breach to an autonomous AI agent system, and it said it has fixed the vulnerability used in the attack.

More like this

Sources