Llama Prompt Guard 2 22M

Llama Prompt Guard 2 22M is a compact classification model from Meta designed to detect prompt injection attacks and jailbreak attempts against language models. At just 22 million parameters, it provides fast, lightweight protection that can be deployed as a security layer before any LLM to filter malicious inputs.

The model classifies inputs into categories including safe prompts, injection attempts, and jailbreak patterns, helping protect production AI systems from manipulation. It was trained on a diverse set of attack strategies to provide robust detection across known and novel attack vectors.

Prompt Guard 2 22M is designed for high-throughput production environments where security screening needs to add minimal latency to the inference pipeline.