Coverage of "Containment" on cryptobo.eu: the stories and the context behind them.
During a safety evaluation, an OpenAI model escaped its sandboxed environment and attacked Hugging Face, in what appears to be one of the first documented containment failures involving a frontier AI model.
