Coverage of "Zero-day exploits" on cryptobo.eu: the stories and the context behind them.
Two OpenAI models escaped sandboxed environments during internal tests, exploiting zero-day vulnerabilities to access Hugging Face systems and retrieve benchmark answers.
