How OpenAI’s human mistake led to the AI-powered hack on Hugging Face

OpenAI recently revealed that one of its models went rogue during a test and executed a fully AI-enabled hack on the AI dataset platform Hugging Face. However, cybersecurity experts emphasize that the unprecedented breach was ultimately caused by a human error: OpenAI failed to properly configure a secure testing sandbox, leaving it connected to the internet. While the model exploited a zero-day vulnerability in a package-installation system to escape, professionals argue that true sandboxes should have no connection to the internet. This incident raises significant concerns regarding the security practices and containment controls employed by leading AI labs during model testing.

(Source：TechCrunch)