newsGoogle News — LLMTrust 62 · AggregatorPublished 12d agoLive · 12d ago
New Research: AI models can give dangerous responses despite output guardrails - The AI Journal
New Research: AI models can give dangerous responses despite output guardrails The AI Journal
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 64%Behind the Refusal: Determining Guardrail Activation via Behavioral Monitoring →
- PossiblePossibly related (embedding) · 53%EvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety Failures →
- PossiblePossibly related (embedding) · 49%Overview of Risk Assessment and Management for Intelligent Systems under the AI Act and Beyond →
- PossiblePossibly related (embedding) · 49%ggwhite/4x →
- PossiblePossibly related (embedding) · 47%Direct Causation in International Humanitarian Law and the Challenge of AI-Mediated Civilian Cyber Operations →
- PossiblePossibly related (embedding) · 58%guardrails-ai/guardrails →
Covers
paperBehind the Refusal: Determining Guardrail Activation via Behavioral MonitoringpaperEvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety FailurespaperOverview of Risk Assessment and Management for Intelligent Systems under the AI Act and Beyondrepoggwhite/4xpaperDirect Causation in International Humanitarian Law and the Challenge of AI-Mediated Civilian Cyber Operations
Covers (incoming)
Related across the graph
paperOverview of Risk Assessment and Management for Intelligent Systems under the AI Act and Beyondrepoggwhite/4xrepoguardrails-ai/guardrailspaperBehind the Refusal: Determining Guardrail Activation via Behavioral MonitoringpaperEvalSafetyGap: A Hybrid Survey and Conceptual Framework for LLM Evaluation-Safety FailurespaperDirect Causation in International Humanitarian Law and the Challenge of AI-Mediated Civilian Cyber Operations