AI SecurityOpenAI's AI Models Find Real Security Flaws in Their Own Systems
A red-teaming exercise using AI models on OpenAI's infrastructure revealed unknown vulnerabilities, including a critical flaw in container software. Training now occurs without internet access.
