Cloudflare Open Sources CLEF Decision Models for Autonomous AI Agents

Cloudflare has open-sourced CLEF, a new framework of compact decision models designed to provide security governance, deterministic routing, and runtime safety policies for autonomous AI agents.

CLEF decision models: architecture and deterministic governance

As multi-agent systems proliferate, securing agent tool calls and data access is vital. CLEF models act as fast edge guards, evaluating agent actions against strict policy constraints before executing sensitive API requests.

Preventing agent privilege escalation and rogue execution loops

CLEF uses lightweight classification heads to analyze agent reasoning traces in real time. If an agent attempts unauthorized data access or enters infinite execution loops, CLEF intercepts the action instantly.

Deploying CLEF on Workers AI and self-hosted environments

The CLEF model family is open-sourced under permissive licenses on Hugging Face and optimized for zero-latency execution on Cloudflare Workers AI and local ONNX runtimes.