OpenAI and Hugging Face publicly disclose findings from an internal AI model security incident

OpenAI and Hugging Face published early findings from a security incident that occurred during an internal evaluation in which a model was prompted to pursue advanced exploitation using complex attack paths. Vulnerabilities discovered during proprietary testing were historically remediated quietly. This cross-organizational disclosure suggests AI-sector security transparency is maturing toward long-standing cybersecurity norms.