Regime-Conditional Verification: Correctness Estimation for Adapting and Monitoring Safety Classifiers

arXiv:2608.14089 · cs.AI, cs.CL, cs.CR, cs.LG · Submitted 2026-08-14 · Read on arXiv

cs.AI, cs.CL, cs.CR, cs.LG

Submitted: 2026-08-14

Updated: 2026-10-06

Comments: 16 pages including technical appendix, 6 figures. Project page and code: https://rcv.tsandoval.com

Project page: https://rcv.tsandoval.com

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Calibration — 47× in this paper · explained in 7 episodes
• Steering — 20× in this paper · explained in 2 episodes
• AUROC — 10× in this paper · explained in 2 episodes
• Verification — 8× in this paper · explained in 2 episodes
• Ablation — 5× in this paper · explained in 1 episode
• Adaptation — 5× in this paper · explained in 1 episode
• Distribution Shift — 5× in this paper · explained in 6 episodes
• Linear probe — 5× in this paper · explained in 2 episodes
• Regime-Conditional Verification — 5× in this paper
• Fine-Tuning — 4× in this paper · explained in 5 episodes
• LLM judge — 4× in this paper · explained in 2 episodes
• Accuracy — 3× in this paper · explained in 1 episode
• Alignment — 3× in this paper · explained in 1 episode
• Benchmark — 2× in this paper · explained in 2 episodes
• Concept Drift — 2× in this paper · explained in 4 episodes
• Conformal Prediction — 2× in this paper · explained in 5 episodes
• Adaptive Monitoring — 1× in this paper · explained in 1 episode
• Failure Modes — 1× in this paper · explained in 3 episodes
• Frozen model — 1× in this paper · explained in 3 episodes
• Generative Model — 1× in this paper · explained in 2 episodes
• Latent Space — 1× in this paper · explained in 7 episodes
• Provenance — 1× in this paper · explained in 4 episodes

Sources

• Calibrating LLM Judges: Linear Probes for Fast and Reliable Uncertainty Estimation
• Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teaming
• DriftGuard: Safety-Aware Multi-Monitor Detection and Selective Adaptation for Evolving Toxicity Moderation
• RepV: Safety-Separable Latent Spaces for Scalable Neurosymbolic Plan Verification
• Safety Cases: How to Justify the Safety of Advanced AI Systems
• The Llama 3 Herd of Models
• WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs
• BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset
• Language Models (Mostly) Know What They Know
• JailbreaksOverTime: Detecting Jailbreak Attacks Under Distribution Shift

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems
• VSAL: A Vision Solver with Adaptive Layouts for Graph Property Detection