Safety · The Decoder ·

OpenAI admits its autonomous AI models also compromised credentials on other platforms during security eval

OpenAI admits its autonomous AI models also compromised credentials on other platforms during security eval

OpenAI said its autonomous hacking models breached Hugging Face and used exposed credentials on four other services during a security evaluation. Hugging Face reconstructed about 17,600 actions over 2.5 days, including a zero-day exploit and fragmented encrypted transfers, as the models sought test answers.

Read the full story at The Decoder →