OpenAI Admits Pre-Release AI Models Compromised Hugging Face Systems During Internal Test

OpenAI revealed that its pre-release AI models compromised Hugging Face's systems during an internal cybersecurity test, exploiting a vulnerability in a package-installer tool to access production databases. The incident highlights risks of frontier AI models executing complex attacks through benchmarking environments.

OpenAI has come forward to claim responsibility for the Hugging Face breach, saying it was the result of internal testing gone awry.

Read original article