New Protection Tools for Llama Developers
summary
Meta releases concrete tools: LlamaGuard 4 is a natively multimodal moderation model that replaces text+image chaining, PromptGuard 2 ships as an 86M and a 22M parameter model for low-latency multilingual prompt-injection detection, and LlamaFirewall provides an orchestration layer that integrates Prompt Guard and Code Shield. CyberSecEval 4 adds AutoPatchBench plus CrowdStrike-backed SOC benchmarks. Also available are a Llama generated audio detector and a Llama 4 speech watermark detector; initial partners include Zendesk, AT&T, and Bell Canada.
AI-written summary · about
chapters (5)
related talks
33mLlamaFirewall: An open source guardrail system for building
37mSecure by Design: Strategies for LLM Adoption in Cloud-Native Environm...
50mSecuring LLM-Powered Applications: Overcoming Security and Privacy Challenges
07mLightning Talk: Serving Guardrail Detectors on Vllm
16mLlama 4
29mBeyond the Hype: Building Trustworthy and Reliable LLM Applications with Guardrails
12mSelfDefend: LLMs Can Defend Themselves against Jailbreaking in...
49mBreaching LLM-Powered Applications: Overcoming Security and Privacy Challenges