Prompt Guard
Meta's tiny 86M classifier purpose-built to catch prompt injections and jailbreaks.
About
Prompt Guard is Meta's open multi-label classifier for detecting direct jailbreaks and indirect prompt injections, fine-tuned from mDeBERTa-v3-base at just 86 million parameters. A v2 release added a 22M variant, and the model is designed as a low-latency (~20ms) tripwire layer in front of LLM applications and agents, later folded into Meta's LlamaFirewall stack.
Limitations: small classifiers can be bypassed by novel attacks — treat as defense-in-depth, not a sole gate; Meta's Llama community license is not OSI-open and carries use restrictions including an acceptable-use policy.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 67; Trending score 68/100 from researcher assessment.