Tag: LLM safety
Browse our exclusive articles!
XL-SafetyBench: Benchmarking LLM Safety & Cultural Sensitivity
Discover XL-SafetyBench, a cross-cultural benchmark testing LLM safety and cultural sensitivity across 10 country-language pairs with advanced metrics.
Policy Invariance: Ensuring Reliable LLM Safety Judges
Discover how policy invariance improves the reliability of LLM safety judges beyond accuracy, ensuring trustworthy AI safety evaluations.
LLM Safety Flaws Revealed by Mathematical Encoding Attacks
Discover how mathematical encoding exposes LLM safety gaps, enabling new attacks with up to 56% success, urging stronger AI safety measures.
Improving Agent Safety with ROME and ARISE Benchmarks
Discover how ROME and ARISE enhance AI agent safety judgment in deceptive scenarios using advanced benchmarks and analogical reasoning.
Persona-Invariant Safety Alignment via Adversarial Self-Play
Discover how adversarial self-play enhances persona-invariant safety alignment in LLMs, reducing jailbreak risks while preserving model performance.