Council warns frontier AI models can accelerate discovery of software vulnerabilities

The convenor briefed members on frontier AI learning models and their implications for software-security work, saying that while models are expensive to run they can be used to discover vulnerabilities at scale. He told the council that "the AI learning models, they're able to find vulnerabilities faster than we're able to fix them because finding them can be automated and leveraged by Frontier models now." The convenor recommended using existing, cheaper models to build secure workflows and noted vendor programs (Anthropic's Project Glasswing and OpenAI governance resources) as useful references.