Tag: agentic AI safety
Browse our exclusive articles!
Refining Safety Rules in CPS Using Grammar-Constrained AI
Explore grammar-constrained refinement of safety rules in cyber-physical systems using AI to ensure syntactic and semantic accuracy for safer operations.
AI Mental Health Training Risks: Clinical Harm Revealed
New research shows AI safety training in mental health can cause harm. Rigorous multi-axis testing is crucial for safe therapeutic AI deployment.
TraceGuard: Black-Box Defense Against Distillation Attacks
Discover TraceGuard, a scalable black-box method protecting AI models from distillation attacks while preserving performance and security.
AI Incident Response: Designing Escalation Criteria & Thresholds
Explore a global framework for AI incident escalation with clear criteria, triggers, and thresholds to enhance international AI safety and response.
Layer-wise Vulnerabilities in LLMs Exposed by Mechanistic Steering
Discover how mechanistic steering reveals layer-specific vulnerabilities in LLMs, enhancing defenses against adversarial jailbreak attacks.