Tag: agentic AI safety
Browse our exclusive articles!
Jailbreak Attacks on Large Reasoning Models Using Semantic Triggers
Explore novel jailbreak attacks on large reasoning models via semantic triggers and psychological framing, revealing key vulnerabilities and defense needs.
Symbolic Guardrails for Safer Domain-Specific AI Agents
Discover how symbolic guardrails improve safety and security in domain-specific AI agents without compromising their utility or performance.
HarmfulSkillBench: Detecting Dangerous Skills in AI Agents
Discover how HarmfulSkillBench identifies and measures harmful skills in AI agents, enhancing safety in large language model ecosystems.
Subliminal Transfer of Unsafe Behaviors in AI Distillation
Explore how unsafe behaviors subliminally transfer in AI agent distillation, revealing risks beyond explicit data sanitation in model training.
IatroBench: Evidence of AI Safety Risks in Medical Advice
IatroBench reveals AI safety measures causing harm by withholding critical medical info, highlighting risks in AI-generated healthcare guidance.