Alignment Verifiability in Large Language Models: Normative Indistinguishability under Behavioral Evaluation

• URL: http://arxiv.org/abs/2602.05656v2
• Date: Fri, 06 Feb 2026 19:05:06 GMT
• Title: Alignment Verifiability in Large Language Models: Normative Indistinguishability under Behavioral Evaluation
• Abstract summary: We study alignment evaluation through the lens of statistical identifiability under partial observability.<n>We formalize the Alignment Verifiability Problem and introduce Normative Indistinguishability.<n>Our results show that behavioral benchmarks provide necessary but insufficient evidence for latent alignment under evaluation awareness.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Behavioral evaluation is the dominant paradigm for assessing alignment in large language models (LLMs). In current practice, observed compliance under finite evaluation protocols is treated as evidence of latent alignment. However, the inference from bounded behavioral evidence to claims about global latent properties is rarely analyzed as an identifiability problem. In this paper, we study alignment evaluation through the lens of statistical identifiability under partial observability. We allow agent policies to condition their behavior on observable signals correlated with the evaluation regime, a phenomenon we term evaluation awareness. Within this framework, we formalize the Alignment Verifiability Problem and introduce Normative Indistinguishability, which arises when distinct latent alignment hypotheses induce identical distributions over evaluator-accessible observations. Our main theoretical contribution is a conditional impossibility result: under finite behavioral evaluation and evaluation-aware policies, observed compliance does not uniquely identify latent alignment, but only membership in an equivalence class of conditionally compliant policies, under explicit assumptions on policy expressivity and observability. We complement the theory with a constructive existence proof using an instruction-tuned LLM (Llama-3.2-3B), demonstrating a conditional policy that is perfectly compliant under explicit evaluation signals yet exhibits degraded identifiability when the same evaluation intent is conveyed implicitly. Together, our results show that behavioral benchmarks provide necessary but insufficient evidence for latent alignment under evaluation awareness.

Related papers

• Beyond Aggregate Scores: Behavioral Correctness Assumptions for Assessing Reference-Based Automatic Evaluation Methods [1.0749048269398973]
  Existing meta-evaluation measures agreement with human judgments or benchmark labels.<n>We introduce behavioral correctness assumptions, a complementary framework for evaluating reference-based automatic evaluation methods.
  arXiv  Detail & Related papers  (2026-09-04T15:43:09Z)
• Position: Evaluation Scores Are Perishable Knowledge Claims [0.0]
  We argue that evaluation scores should be treated as claims with three properties: formality, scope, and validity windows.<n>We illustrate the cost of mean aggregation on the public HELM leaderboard.
  arXiv  Detail & Related papers  (2026-07-28T18:50:39Z)
• From Holistic Evaluation to Structured Criteria: Rubrics Across the Evolving LLM Landscape [79.30826980815927]
  rubrics are explicit criteria sets that transform complex quality judgments into structured and actionable standards.<n>We systematically organize existing rubric designs, examine their construction and optimization, and analyze their role across evaluation and training.
  arXiv  Detail & Related papers  (2026-06-07T13:34:55Z)
• Models That Know How Evaluations Are Designed Score Safer [38.21092181000792]
  We investigate evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations.<n>Similar to contamination dataset, where benchmark exposure leads to higher performance through memorization, we hypothesize that models trained on texts describing evaluation practices may implicitly learn to recognize and respond to evaluation-like contexts.<n>Our results demonstrate that evaluation meta-knowledge may inflate safety benchmark performance, introducing a novel confounder that is independent of explicit memorization or verbalized evaluation awareness.
  arXiv  Detail & Related papers  (2026-05-27T15:11:35Z)
• In-Situ Behavioral Evaluation for LLM Fairness, Not Standardized-Test Scores [20.623498601506974]
  We show that the standardized-test paradigm can be structurally unreliable.<n>We develop MAC-Fairness, a multi-agent conversational framework that embeds controlled variation factors into multi-round dialogue.<n>In-situ behavioral evaluation reveals stable, model-specific behavioral signatures that could generalize across benchmarks differing in fairness targets and evaluation methodologies.
  arXiv  Detail & Related papers  (2026-04-21T18:38:50Z)
• On the Structural Non-Preservation of Epistemic Behaviour under Policy Transformation [51.56484100374058]
  We formalise such information-conditioned interaction patterns as behavioural dependency.<n>This induces a probe-relative notion of $$-behavioural equivalence and a within-policy behavioural distance.<n>Results identify structural conditions under which probe-conditioned behavioural separation is not preserved under common policy transformations.
  arXiv  Detail & Related papers  (2026-02-24T22:55:21Z)
• Causality is Key for Interpretability Claims to Generalise [35.833847356014154]
  Interpretability research on large language models (LLMs) has yielded important insights into model behaviour.<n> recurring pitfalls persist: findings that do not generalise, and causal interpretations that outrun the evidence.<n>Pearl's causal hierarchy clarifies what an interpretability study can justify.
  arXiv  Detail & Related papers  (2026-02-18T18:45:04Z)
• The Validity of Coreference-based Evaluations of Natural Language Understanding [3.505146496638911]
  I analyze standard coreference evaluations and show that their design often leads to non-generalizable conclusions.<n>I propose and implement a novel evaluation focused on testing systems' ability to infer the relative plausibility of events.
  arXiv  Detail & Related papers  (2026-02-18T05:49:28Z)
• Operational Noncommutativity in Sequential Metacognitive Judgments [0.0]
  We show that metacognitive evaluations are modeled as state-transforming operations acting on an internal state space with probabilistic readouts.<n>We also outline a behavioral paradigm involving sequential confidence, error-likelihood, and feeling-of-knowing judgments following a perceptual decision.
  arXiv  Detail & Related papers  (2026-02-15T08:15:48Z)
• When Evaluation Becomes a Side Channel: Regime Leakage and Structural Mitigations for Alignment Assessment [0.0]
  Safety evaluation for advanced AI systems assumes that behavior observed under evaluation predicts behavior in deployment.<n>We recast alignment evaluation as a problem of information flow under partial observability.<n>We study regime-blind mechanisms, training-time interventions that restrict access to regime cues.
  arXiv  Detail & Related papers  (2026-02-09T10:00:24Z)
• Alignment-Aware Model Adaptation via Feedback-Guided Optimization [27.93864970404945]
  Fine-tuning is the primary mechanism for adapting foundation models to downstream tasks.<n>We propose an alignment-aware fine-tuning framework that integrates feedback from an external alignment signal through policy-gradient-based regularization.
  arXiv  Detail & Related papers  (2026-02-02T16:03:16Z)
• Probability-Entropy Calibration: An Elastic Indicator for Adaptive Fine-tuning [55.2818264614932]
  RankTuner introduces a probability--entropy calibration signal, the Relative Rank Indicator, which compares the rank of the ground-truth token with its expected rank under the prediction distribution.<n>The inverse indicator is used as a token-wise Relative Scale to reweight the fine-tuning objective, focusing updates on truly under-learned tokens.
  arXiv  Detail & Related papers  (2026-02-02T07:27:19Z)
• Character as a Latent Variable in Large Language Models: A Mechanistic Account of Emergent Misalignment and Conditional Safety Failures [70.48661957773449]
  Emergent Misalignment refers to a failure mode in which fine-tuning large language models on narrowly scoped data induces broadly misaligned behavior.<n>Across multiple domains and model families, we find that fine-tuning models on data exhibiting specific character-level dispositions induces substantially stronger and more transferable misalignment than incorrect-advice fine-tuning.
  arXiv  Detail & Related papers  (2026-01-30T15:28:42Z)
• Logical Consistency Between Disagreeing Experts and Its Role in AI Safety [0.0]
  Two experts disagree on a test, we may conclude both cannot be 100 per cent correct. But if they completely agree, no possible evaluation can be excluded.<n>This asymmetry in the utility of agreements versus disagreements is explored here by formalizing a logic of unsupervised evaluation for classifiers.
  arXiv  Detail & Related papers  (2025-10-01T12:30:01Z)
• Reference-Free Rating of LLM Responses via Latent Information [53.463883683503106]
  We study the common practice of asking a judge model to assign Likert-scale scores to free-text responses.<n>We then propose and evaluate Latent Judges, which derive scalar ratings from internal model signals.<n>Across a broad suite of pairwise and single-rating benchmarks, latent methods match or surpass standard prompting.
  arXiv  Detail & Related papers  (2025-09-29T12:15:52Z)
• Alignment Quality Index (AQI) : Beyond Refusals: AQI as an Intrinsic Alignment Diagnostic via Latent Geometry, Cluster Divergence, and Layer wise Pooled Representations [8.454242629883488]
  We introduce the Alignment Quality Index (AQI) to assess large language models' alignment.<n>AQI captures clustering quality to detect hidden misalignments and jailbreak risks, even when outputs appear compliant.<n>We also propose the LITMUS dataset to facilitate robust evaluation under challenging conditions.
  arXiv  Detail & Related papers  (2025-06-16T18:22:28Z)
• Understanding challenges to the interpretation of disaggregated evaluations of algorithmic fairness [49.35494016290887]
  We show that equal performance across subgroups is an unreliable measure of fairness when data are representative of relevant populations but reflective of real-world disparities.<n>Our framework suggests complementing disaggregated evaluations with explicit causal assumptions and analysis to control for confounding and distribution shift.
  arXiv  Detail & Related papers  (2025-06-04T17:40:31Z)
• Retrieval is Not Enough: Enhancing RAG Reasoning through Test-Time Critique and Optimization [58.390885294401066]
  Retrieval-augmented generation (RAG) has become a widely adopted paradigm for enabling knowledge-grounded large language models (LLMs)<n>RAG pipelines often fail to ensure that model reasoning remains consistent with the evidence retrieved, leading to factual inconsistencies or unsupported conclusions.<n>We propose AlignRAG, a novel iterative framework grounded in Critique-Driven Alignment (CDA)<n>We introduce AlignRAG-auto, an autonomous variant that dynamically terminates refinement, removing the need to pre-specify the number of critique iterations.
  arXiv  Detail & Related papers  (2025-04-21T04:56:47Z)
• Internal Incoherency Scores for Constraint-based Causal Discovery Algorithms [12.524536193679124]
  We propose internal coherency scores that allow testing for assumption violations and finite sample errors.<n>We illustrate our coherency scores on the PC algorithm with simulated and real-world datasets.
  arXiv  Detail & Related papers  (2025-02-20T16:44:54Z)
• Estimating Commonsense Plausibility through Semantic Shifts [66.06254418551737]
  We propose ComPaSS, a novel discriminative framework that quantifies commonsense plausibility by measuring semantic shifts.<n> Evaluations on two types of fine-grained commonsense plausibility estimation tasks show that ComPaSS consistently outperforms baselines.
  arXiv  Detail & Related papers  (2025-02-19T06:31:06Z)
• The simulation of judgment in LLMs [32.57692724251287]
  Large Language Models (LLMs) are increasingly embedded in evaluative processes, from information filtering to assessing and addressing knowledge gaps through explanation and credibility judgments.<n>This raises the need to examine how such evaluations are built, what assumptions they rely on, and how their strategies diverge from those of humans.<n>We benchmark six LLMs against expert ratings--NewsGuard and Media Bias/Fact Check--and against human judgments collected through a controlled experiment.
  arXiv  Detail & Related papers  (2025-02-06T18:52:10Z)
• Mitigating LLM Hallucinations via Conformal Abstention [70.83870602967625]
  We develop a principled procedure for determining when a large language model should abstain from responding in a general domain. We leverage conformal prediction techniques to develop an abstention procedure that benefits from rigorous theoretical guarantees on the hallucination rate (error rate) Experimentally, our resulting conformal abstention method reliably bounds the hallucination rate on various closed-book, open-domain generative question answering datasets.
  arXiv  Detail & Related papers  (2024-04-04T11:32:03Z)
• Matrix Estimation for Offline Reinforcement Learning with Low-Rank Structure [10.968373699696455]
  We consider offline Reinforcement Learning (RL), where the agent does not interact with the environment and must rely on offline data collected using a behavior policy. Previous works provide policy evaluation guarantees when the target policy to be evaluated is covered by the behavior policy. We propose an offline policy evaluation algorithm that leverages the low-rank structure to estimate the values of uncovered state-action pairs.
  arXiv  Detail & Related papers  (2023-05-24T23:49:06Z)
• Hallucinated Adversarial Control for Conservative Offline Policy Evaluation [64.94009515033984]
  We study the problem of conservative off-policy evaluation (COPE) where given an offline dataset of environment interactions, we seek to obtain a (tight) lower bound on a policy's performance. We introduce HAMBO, which builds on an uncertainty-aware learned model of the transition dynamics. We prove that the resulting COPE estimates are valid lower bounds, and, under regularity conditions, show their convergence to the true expected return.
  arXiv  Detail & Related papers  (2023-03-02T08:57:35Z)
• Proximal Reinforcement Learning: Efficient Off-Policy Evaluation in Partially Observed Markov Decision Processes [65.91730154730905]
  In applications of offline reinforcement learning to observational data, such as in healthcare or education, a general concern is that observed actions might be affected by unobserved factors. Here we tackle this by considering off-policy evaluation in a partially observed Markov decision process (POMDP) We extend the framework of proximal causal inference to our POMDP setting, providing a variety of settings where identification is made possible.
  arXiv  Detail & Related papers  (2021-10-28T17:46:14Z)
• Deconfounding Scores: Feature Representations for Causal Effect Estimation with Weak Overlap [140.98628848491146]
  We introduce deconfounding scores, which induce better overlap without biasing the target of estimation. We show that deconfounding scores satisfy a zero-covariance condition that is identifiable in observed data. In particular, we show that this technique could be an attractive alternative to standard regularizations.
  arXiv  Detail & Related papers  (2021-04-12T18:50:11Z)
• Confounding-Robust Policy Evaluation in Infinite-Horizon Reinforcement Learning [70.01650994156797]
  Off- evaluation of sequential decision policies from observational data is necessary in batch reinforcement learning such as education healthcare. We develop an approach that estimates the bounds of a given policy. We prove convergence to the sharp bounds as we collect more confounded data.
  arXiv  Detail & Related papers  (2020-02-11T16:18:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.