LlamaFirewall: An open source guardrail system for building
summary
LlamaFirewall is an open source guardrail framework that ships three concrete tools: PromptGuard 86M/22M models as a lightweight jailbreak classifier with real‑time latency and over 1M downloads; Agent Alignment Checks that audit chain‑of‑thought traces and catch about 80–90% of misaligned tool actions at roughly 2–4% false positives (baseline attack success ≈48%); and CodeShield, a fast online static analyzer for LLM‑generated code. Scanners plug into an orchestration layer and run in production at Meta.
AI-written summary · about
chapters (7)
- 0:00AI agent security intro
- 7:27LlamaFirewall framework
- 12:40Prompt attack threats
- 17:59Jailbreak detection
- 20:24Agent misalignment check
- 23:55AI security closing
- 27:45Q&A
related talks
11mNew Protection Tools for Llama Developers
37mSecure by Design: Strategies for LLM Adoption in Cloud-Native Environm...
07mLightning Talk: Serving Guardrail Detectors on Vllm
26mOne guardrail is not enough: Defense-in-depth against LLM jailbreaks
12mSelfDefend: LLMs Can Defend Themselves against Jailbreaking in...
46mGuarding the LLM Galaxy: Security, Privacy, and Guardrails in the...
29mBeyond the Hype: Building Trustworthy and Reliable LLM Applications with Guardrails
49mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges