cs.AINov 6, 2026

BMMS: Policy-Conditioned Measurement of Behavioral Compliance in Language Models

Authors: Anthony Rhodes, Gadi Singer

Organizations: Confidential Core AI

Abstract

Behavioral evaluation of language models fixes its dimensions in advance: hallucination, sycophancy, toxicity, over-refusal. An operator in a regulated industry needs a different question answered: does a system do what a particular rulebook requires? We introduce the Behavioral Misalignment Metric Suite (BMMS), which measures policy-conditioned misalignment by taking a policy corpus as an argument and returning a compliance profile over it, for any published policy and any system that answers in natural language. The Compliance Attainment Score (CAS) measures severity-weighted attainment over the obligations that bind the system, and the Compliance Robustness Margin (CRM) the share of that attainment surviving graded adversarial pressure; Behavioral Invariance (BIV) and the Behavioral Disposition Index (BDI) are specified and released with the suite. Dimensions enter only by passing a published admission test whose binding criterion is attribution: a quantity that moves when the judge changes, holding the target fixed, is measuring the instrument. Applying it excluded one of the quantities the suite itself reported. Across 1,003 obligations from eleven documents in five registers, the admission criterion holds 94.4% precision against blinded human judgment on vendor specifications and 50.0% on analytical guidance, extraction fidelity is 94.6%, and re-scoring under judges from two other providers leaves target ordering invariant while the absolute score of a single target spans up to 0.22. We apply it to the two most detailed published behavioral specifications. Own-specification advantage is not resolved from zero once document difficulty is subtracted, but absolute self-compliance is another matter: GPT-4o fails 70.7% of the hard constraints binding it in the Model Spec published by OpenAI, where the two Claude models fail 0.0% and 9.1% of theirs.

Figures & tables

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Oct 5, 2026

The Pushback Paradox: A Two-Probe Diagnostic for Language Model Compliance

Are language models compliant with user instructions? A model that always complies can be stopped but also exploited, while one that always resists can be neither exploited nor stopped. We contribute an open two-probe benchmark that can place any language model on this spectrum. In the active probe, a user instructs the model to act and accept a lower payoff, which measures exploitability. In the passive probe, the user instructs it to wait and give up a higher payoff, which measures stoppability. The two compliance rates combine into a compliance index κκ. Applied to twelve language models, the benchmark shows that seven mostly follow the instruction in both probes and justify their action by pointing to the instruction. Only Claude Sonnet-4.6 and Claude Opus-4.7 can be stopped without being exploitable, Claude Opus-4.6 and GPT-5-mini resist both instructions, and no model is exploitable but unstoppable. Knowing where a language model sits on the compliance index κκ matters for human operators and for multi-agent systems, whether distributed or orchestrated.
Aug 17, 2026

What Do Compliance Detectors Read? An Audit of Activation Probes and Guard Models

Regulatory compliance monitoring in deployed language models is increasingly implemented as a legal and audit control, checking model outputs against written rules spanning data protection, healthcare, financial regulation, and platform policy. Such monitoring is meaningful only if a detector's verdict depends on the stated rule rather than on surface features of the scenario. We show this condition fails across the current class of compliance detectors, a failure we call rule blindness. Deleting, permuting, or substituting the governing rule leaves detection accuracy unchanged for every guard and activation probe we test, including a policy-conditioned guard that correctly cites the governing clause yet barely changes its verdict when that clause is swapped for its permissive counterpart. A purpose-built benchmark crossing two rules with two scenarios, so that neither alone predicts the label, confirms the failure under a design no prior benchmark rules out, and shows that step by step reasoning, not any fast detector we test, is what escapes it. Auditing at scale requires a retraining-free detector, so we introduce the Internal Compliance Score (ICS): a training-free activation readout calibrated from ten labelled pairs and scored by a single projection. We hold ICS to the same scrutiny as the guards it audits: a pre-registered criterion for beating trivial baselines is not met, and a bag-of-words model matches its pooled generalisation exactly. It remains useful because it is inexpensive, letting us audit four deployed guard models, an 8B zero-shot judge, and thirteen benchmarks, and it raises the mechanically verified pass rate when used to rank candidate responses, though an adaptive white-box attack removes this gain. We release the counterfactual protocol and crossed-rule benchmark so rule blindness can be tested in future probe and guard claims.
Feb 24, 2026

Pressure Reveals Character: Behavioural Alignment Evaluation at Depth

Evaluating alignment in language models requires testing how they behave under realistic pressure, not just what they claim they would do. While alignment failures increasingly cause real-world harm, comprehensive evaluation frameworks with realistic multi-turn scenarios remain lacking. We introduce an alignment benchmark spanning 904 scenarios across six categories -- Honesty, Safety, Non-Manipulation, Robustness, Corrigibility, and Scheming -- validated as realistic by human raters. Our scenarios place models under conflicting instructions, simulated tool access, and multi-turn escalation to reveal behavioural tendencies that single-turn evaluations miss. Evaluating 24 frontier models using LLM judges validated against human annotations, we find that even top-performing models exhibit gaps in specific categories, while the majority of models show consistent weaknesses across the board. Factor analysis reveals that alignment behaves as a unified construct (analogous to the g-factor in cognitive research) with models scoring high on one category tending to score high on others. We publicly release the benchmark and an interactive leaderboard to support ongoing evaluation, with plans to expand scenarios in areas where we observe persistent weaknesses and to add new models as they are released.