Cloudflare details its internal security architecture built to defend against AI-accelerated cyber attacks, using its own products as 'customer zero'. The post covers three key threat vectors from frontier AI models: faster vulnerability discovery, high-volume adaptive exploit generation, and blast radius after exploitation. The architecture layers WAF with ML-based scoring, API Shield with positive security models, Bot Management, Zero Trust Network Access, IdP Federation, MCP Server Portal, and AI Gateway. Cloudflare's WAF Attack Score assigns risk ratings per request based on attack shape similarity rather than known signatures, enabling detection of novel exploits. The post also describes continuous red team testing both at the perimeter and inside the environment, and recommends starting with inspection in front of public apps, defining valid API traffic, bot detection, and identity-gated internal tools.
