Tag: LLM safety
Browse our exclusive articles!
TraceSafe: Assessing LLM Guardrails in Multi-Step Tool Use
Explore TraceSafe-Bench, the first benchmark evaluating LLM guardrails' effectiveness in multi-step tool-calling trajectories for enhanced AI safety.
Why Prompt Injection Defense Wrappers Often Fail
Explore the defense trilemma in AI security and why prompt injection defense wrappers struggle to balance safety, utility, and continuity.
RETINA-SAFE & ECRT: Reducing Hallucination Risks in Medical LLMs
Discover how RETINA-SAFE and ECRT frameworks improve hallucination risk triage in medical LLMs, enhancing patient safety with evidence-based AI decisions.
Scalable Safety Evaluations of LLMs for Psychosis Support
Explore scalable, clinically-validated safety evaluations of LLM responses for users with psychosis, enhancing mental health AI support systems.
Reactivating Hidden Safety in Post-Trained Large Language Models
Discover how SafeReAct restores suppressed safety mechanisms in post-trained LLMs, enhancing security without losing reasoning power.