From Probe Scores to Alarm Policies: Operational Validity of Activation Monitors for Language-Model Agents
cs.CL, cs.AI
Submitted: 2026-10-03
Updated: 2026-10-03
Comments: 12 pages, 1 figure. Accepted at the Eighth International Conference on Distributed Artificial Intelligence (DAI 2026)
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Sources
- From Confident Closing to Silent Failure: Characterizing False Success in LLM Agents
- Conformal Risk Control
- Refusal Before Decoding: Detecting and Exploiting Refusal Signals in Intermediate LLM Activations
- Do Activation Monitors Survive Model Updates? Benchmarking, Predicting, and Repairing Activation-Monitor Staleness
- CORA: Conformal Risk-Controlled Agents for Safeguarded Mobile GUI Automation
- Evaluation Cards: An Interpretive Layer for AI Evaluation Reporting
- AI Control: Improving Safety Despite Intentional Subversion
- Unifying Evaluation of Machine Learning Safety Monitors
- PrefixGuard: Online Failure Warning and Trace-Grounded Diagnosis for LLM Agents
- The Impact of Off-Policy Training Data on Probe Generalisation
- A Structured Approach to Safety Case Construction for AI Systems
- TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling
- Do Activation Verbalization Methods Convey Privileged Information?
- When AUC 0.998 Is Not Enough: A Candidate Evaluation Protocol for Hidden-State Probes of Indirect Prompt Injection in Multimodal Computer-Use Agents
- Agent MechSuits: Mechanistic Subspace Safety Steering for Multi-Turn CLI Agents
- SLEIGHT-Bench: A Benchmark of Evasion Attacks Against Agent Monitors
- Benchmarking Deception Probes via Black-to-White Performance Boosts
- Online Safety Monitoring for LLMs
- Accurate Failure Prediction in Agents Does Not Imply Effective Failure Prevention
- PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis
Related papers
- Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
- Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
- Subliminal Steering: Stronger Encoding of Hidden Signals
- MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
- The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
- Untangling the Mechanisms of Misleading Context in Medical Question Answering