Exploit Before Discovery: Frontier AI and Cybersecurity

The technical cybersecurity aspect of frontier models is especially pertinent right now. OpenAI’s GPT-5.5 and Anthropic’s Mythos Preview both possess incredibly advanced offensive cyber security capabilities with wildly different release strategies - Mythos restricted to vetted software companies and cyber defence, while GPT-5.5 is publicly available with model-based guardrails. With capabilities proliferating from frontier models to more widely available models, what is the role of government cyber agencies in structuring oversight and guiding industry? Without it, government, industry and small businesses will be left vulnerable.