AI Robustness Evaluation System
EvoSynth is a SOTA automated LLM red-teaming framework that evolves executable, code-level jailbreak attack methods using multi-agent synthesis, evolutionary search, and self-correction.
[EMNLP 2024] Holistic Automated Red Teaming for Large Language Models through Top-Down Test Case Generation and Multi-turn Interaction
Meet Claudia: An adversarial AI agent for Project Sentinel that relentlessly tests, breaks, and fixes machine-generated code before a human has to review it.
Nexus is a multi-agent, safety-first Continuous Automated Red Teaming (CART) and Breach & Attack Simulation (BAS) platform. It continuously discovers assets, builds attack graphs, and executes non-destructive proof-of-concept exploits while autonomously learning and adapting to the target environment's defensive posture.
RedBot is an open-source, intelligent adversarial security evaluation platform for LLM systems. It automatically discovers, exploits, and maps vulnerabilities in any LLM-powered application by simulating a skilled human red-teamer.
To associate your repository with the automated-red-teaming topic, visit your repo's landing page and select "manage topics."