Beyond the ban: A better way to secure generative AI applications

Cloudflare introduces AI prompt protection, a DLP capability in Cloudflare One that provides visibility and control over prompts and responses in web-based GenAI tools (e.g., ChatGPT, Gemini, Claude, Perplexity). It detects and classifies prompts by content and intent (e.g., PII, credentials, source code, jailbreak/malicious code), enforces granular guardrails based on user identity and application actions, and securely logs conversations with customer-managed encryption. Under the hood, it leverages operation mapping, protocol-aware DLP parsing (including streaming responses), and a multi-model classification pipeline (Presidio, Promptguard2, Llama3-70B, Vectorize embeddings on Workers AI). The feature is in beta, with plans to broaden app support, enhance workflows, and deepen integrations.