🛡️AutoControl Arena: Synthesizing Executable Test Environments for Frontier AI Risk Evaluation

Ferret: Faster and Effective Automated Red Teaming with Reward-Based Scoring Technique

[EMNLP 2024] Holistic Automated Red Teaming for Large Language Models through Top-Down Test Case Generation and Multi-turn Interaction

Safety Mirage: How Spurious Correlations Undermine VLM Safety Fine-tuning

[CVPR'26] SafeGRPO: Self-Rewarded Multimodal Safety Alignment via Rule-Governed Policy Optimization

[ICML 2025] Vulnerability-Aware Alignment: Mitigating Uneven Forgetting in Harmful Fine-Tuning

Official Implementation of "PSAlign: Personalized Safety Alignment for Text-to-Image Diffusion Models"

AgentAlign: Navigating Safety Alignment in the Shift from Informative to Agentic Large Language Models

This is the official implementation (code, data) of the paper "MOSSBench: Is Your Multimodal Language Model Oversensitive to Safe Queries?""

SEALGuard: Safeguarding the Multilingual Conversations in Southeast Asian Languages for LLM Software Systems

Code for the paper: "Co-Evolutionary Multi-Modal Alignment via Structured Adversarial Evolution".

Adversarial testing of LLMs on constraint satisfaction deadlocks

A PID-Lagrangian based Reinforcement Unlearning framework

Risk-Aware Introspective RAG (RAI-RAG) is a safety-aligned RAG framework integrating introspective reasoning, risk-aware retrieval gating, and secure evidence filtering to build trustworthy, robust, and secure LLM and agentic AI systems.

Quantifying the "Safety Half-Life" of LLMs: A framework to measure how safety alignment degrades and susceptibility to jailbreaks increases as context length grows