El lado del mal - Llama Protections: LlamaFirewall con PromptGuard 2, LlamaGuard 4, AlignmentCheck, CodeShield AutoPatchBench & CyberSecEval 4 elladodelmal.com/2025/04/lla… #Llama #LLM #Hardening #Ciberseguridad #PromptInjection #Jailbreak #Llama4 #CodeShield #IA #OpenSource #IA

AIによるセキュリティ修正のベンチマークAutoPatchBenchをMeta社のエンジニア陣が提案。同社の防衛用途向けAI能力評価ベンチマーク群CyberSecEval 4の一部。実際にファジングで発見されたC/C での脆弱性136件とARVOデータセットからの修正からなる。 engineering.fb.com/2025/04/2…

Introducing AutoPatchBench: A Benchmark for AI-Powered Security Fixes engineering.fb.com/2025/04/2… #RSAC

Meta has released new #AI security tools including #LlamaFirewall and #AutoPatchBench to address rising concerns around prompt injections, insecure code, and fake audio threats. news9live.com/technology/art…

🔥 Meta just dropped a firewall for AI. LlamaFirewall is open-source—and built to stop jailbreaks, prompt injections, and insecure code in real time. It’s modular. It’s fast. It’s made for the LLM era. 🛡️ Also out: 🔹 CyberSecEval 4 with AutoPatchBench to test AI-powered vuln fixes 🔹 Llama for Defenders to help fight scams, fraud & phishing 🔹 Private Processing to run AI features without leaking user data 🔗 Full details here: thehackernews.com/2025/04/me…

Meta launched open source tools to support the open source GenAI security ecosystem 1. LlamaFirewall; a security-first guardrail framework for mitigating agentic prompt injection, misalignment, and insecure coding risks - meta-llama.github.io/PurpleL… 2. Introducing AutoPatchBench: A Benchmark for AI-Powered Security Fixes - engineering.fb.com/2025/04/2… 3. ClassifyIt: Google Workspace Bulk Content Classification - github.com/meta-llama/Purple… 4. CodeShield - Shield against LLM generated insecure code - github.com/meta-llama/Purple… By @AIatMeta @Meta #GenAISecurity #OpenSourceAI #PurpleLlama #LlamaFirewall #AutoPatchBench #CodeShield #ClassifyIt #SecureAI #LLMSecurity #MetaAI

- AutoPatchBench; challenging, grounded evals to help the community build automatic AI security vuln fixing engineering.fb.com/2025/04/2… - A sensitive document classification framework that makes it easy to apply LLMs to preventing sensitive data exfil github.com/meta-llama/Purple… ...

We are introducing AutoPatchBench, a benchmark for the automated repair of vulnerabilities identified through fuzzing. By providing a standardized benchmark, AutoPatchBench enables researchers and practitioners to objectively evaluate and compare the effectiveness of various AI program repair systems. This initiative facilitates the development of more robust security solutions, and also encourages collaboration within the community to address the critical challenge of software vulnerability repair. Read more about AutoPatchBench and try it now on GitHub! engineering.fb.com/2025/04/2…