… Our framework SCOUT Scalable and Controllable Outcome-prediction for Uncertainty-aware Triage makes this decision dynamic by predicting each detector's per-sample reliability and latency from how it behaved on similar past inputs, and exposes a single safety-utility threshold to the operator where … …
… However, their safety has received relatively limited attention. …
… The following papers were recommended by the Semantic Scholar API Few Tokens, Big Leverage: Preserving Safety Alignment by Constraining Safety Tokens during Fine-tuning 2026 SafeAnchor: Preventing Cumulative Safety Erosion in Continual Domain Adaptation of Large Language Models 2026 Risk-Adjusted H… …
… We further introduce ICU -Evo to study structured-memory agents that improves long-horizon reasoning but does not fully eliminate safety failures. …
… This study developed and preliminarily evaluated a domain-specific audit framework for medical research agent skills , with a focus on reliability against expert review . …
… The key finding: capital-agent reliability is an operating-layer problem. The largest reliability gains came from prompt compilation, typed controls, policy validation, execution guards, memory semantics, and full instruction-to-settlement observability. …
… View arXiv page View PDF Add to collection Community For questions or model-evaluation requests, contact guijin.son@snu.ac.kr . objective evaluation을 하기에는 class imbalance가 너무 큰게 아닌게 싶네요. the way soohak treats refusal as a first-class signal is a clever move, highlighting a real frontier beyond pure… …
… The following papers were recommended by the Semantic Scholar API AgentTrap: Measuring Runtime Trust Failures in Third-Party Agent Skills 2026 Defenses&Enablers For Skill Injection Attacks on Terminal Based Agents 2026 SkillHarm: Lifecycle-Aware Skill-Based Attacks via Automated Construction 2026 F… …
… The following papers were recommended by the Semantic Scholar API SkillSafetyBench: Evaluating Agent Safety under Skill-Facing Attack Surfaces 2026 AgentTrap: Measuring Runtime Trust Failures in Third-Party Agent Skills 2026 POISE: Position-Aware Undetectable Skill Injection on LLM Agents 2026 Plan… …
… The following papers were recommended by the Semantic Scholar API Claw-Eval: Toward Trustworthy Evaluation of Autonomous Agents 2026 ClawEnvKit: Automatic Environment Generation for Claw-Like Agents 2026 One-Eval: An Agentic System for Automated and Traceable LLM Evaluation 2026 GTA-2: Benchmarking… …