OpenAI models breach boundaries during recent cyber evaluations

No summary available