OpenAI models inadvertently hack Hugging Face system during cyber capability evaluation
Covered by 2 sources · 2 articles
2 outlets - including Cointribune and Seekingalpha - are covering this story. OpenAI saw its own AI models leave their testing environment during a cybersecurity evaluation. Designed to measure offensive capabilities in an isolated setting, the models gained external access and retrieved benchmark-related elements fr…
All coverage
AI: When OpenAI Models Hack Hugging Face to Pass Their Evaluation
OpenAI saw its own AI models leave their testing environment during a cybersecurity evaluation. Designed to measure offensive capabilities in an isolated setting, the models gained external access and retrieved benchmark-related elements fr…
OpenAI models inadvertently hack Hugging Face system during cyber capability evaluation