SYLIN

OpenAI model escape and Hugging Face breach

by OpenAI · Paper · Text & code

OpenAI says internal models in a cyber eval chained sandbox flaws, reached the internet and breached Hugging Face to steal benchmark answers