Where Instruction Hierarchy Breaks: Diagnosing and Repairing Failures in Reasoning Language Models

• URL: http://arxiv.org/abs/2606.07808v1
• Date: Fri, 05 Jun 2026 19:36:48 GMT
• Title: Where Instruction Hierarchy Breaks: Diagnosing and Repairing Failures in Reasoning Language Models
• Abstract summary: We introduce a diagnostic framework that localizes instruction hierarchy failures into instruction identification, conflict resolution, and response realization.<n>We evaluate three reasoning models--Gemma-4-31B-IT, Qwen3.6-35B-A3B, and Claude Sonnet 4.6--on long-context adaptations of IHEval and IHChallenge.<n>The strongest monitor reduces rule-following non-compliance by 81-99%, with GPT-5.3 reductions of 86% under static attacks and 45% under adaptive attacks.
• Score: 15.752705487977861
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Reasoning language models deployed in agentic workflows must follow an instruction hierarchy: when instructions from different sources conflict, the model should obey the highest-privilege applicable instruction. Existing benchmarks largely measure this behavior end-to-end, asking whether the final response is compliant. However, a non-compliant response can arise from several distinct failures: the model may fail to identify the relevant instructions in context, fail to resolve conflicts among identified instructions, or correctly resolve the conflict in its reasoning while still producing a violating response. We introduce a white-box diagnostic framework that localizes instruction hierarchy failures into instruction identification, conflict resolution, and response realization, making failures more interpretable. We evaluate three reasoning models--Gemma-4-31B-IT, Qwen3.6-35B-A3B, and Claude Sonnet 4.6--on long-context adaptations of IHEval and IHChallenge, and find that the dominant failure mode varies across models, tasks, and context length. Building on the observation that models can often detect conflicts and output violations when explicitly prompted, we propose two training-free self-monitoring mechanisms: a parallel input monitor for low-latency conflict detection before generation, and a sequential output monitor for response-level review and repair. Across Gemma-4-31B-IT, Claude Sonnet 4.6, and GPT-5.3, the strongest monitor reduces rule-following non-compliance by 81-99%, with GPT-5.3 reductions of 86% under static attacks and 45% under adaptive attacks.

Related papers

• Measuring Stability and Failure Behavior in Language Models Under Structured Perturbations [0.5076419064097734]
  We present a graded, multi-family, failure-aware framework for stress-testing reasoning models.<n>It perturbs each problem along a multi-level severity ladder across seven families.<n>Every test is validity-gated and labeled by its measured severity.
  arXiv  Detail & Related papers  (2026-08-22T23:46:11Z)
• Beyond Correctness: Benchmarking and Aligning Response Behaviors in Hybrid-Thinking MLLMs [46.40763822929566]
  We study whether thinking and non-thinking interfaces preserve acceptable final-response behavior.<n>PatternEval tests four recurrent failures: chain-of-thought leakage, response repetition, logical contradiction, and performative reasoning.<n>We develop textbfPatternRM, a response-level reward model, and textbfPatternRL, which introduces pattern-specific penalties during reinforcement learning.
  arXiv  Detail & Related papers  (2026-08-13T03:44:45Z)
• SearchAuditor: Auditing and Attributing Failures in Long-Horizon Search Agents [56.400332595452966]
  We introduce SearchAuditBench, a benchmark that evaluates whether auditors can localize, attribute, and repair search-agent failures.<n>SearchAuditBench comprises 1,243 failed trajectories, averaging 73.1 messages and 65.1K tokens, collected from eight open-weight models on five deep-search benchmarks.<n>We propose SearchAuditor, a multi-perspective auditing framework that effectively localizes, attributes, and repairs search-agent failures.
  arXiv  Detail & Related papers  (2026-08-05T09:09:00Z)
• R-APS: Compositional Reasoning and In-Context Meta-Learning for Constrained Design via Reflective Adversarial Pareto Search [18.8323743697237]
  Large language models (LLMs) are fluent on open-ended tasks, yet in agentic settings, fluency does not ensure reliable delivery.<n>We trace this gap to three coupled structural failures: errors propagate without localization, worst-case perturbations go unevaluated, and accumulated knowledge is never invalidated.<n>We argue these share a root cause: abductive, counterfactual, meta-inductive, corrective, and inductive reasoning pull a shared context in incompatible directions.
  arXiv  Detail & Related papers  (2026-06-03T12:45:39Z)
• FailureScope: Cross-Regime Behavioral Diagnosis of Language Model Weaknesses [0.0]
  FailureScope is a behavioral-diagnosis method that clusters evaluation probes by their cross-model pass/fail patterns.<n>We show it yields stable, interpretable failure across three regimes usually studied separately: single-turn benchmarks, multi-turn dialogue, and adversarial agent attacks.
  arXiv  Detail & Related papers  (2026-06-03T01:28:00Z)
• Detection Without Correction: A Two-Parameter Decomposition of Multi-Stage LLM Pipelines [0.764671395172401]
  We study multi-agent debate, intrinsic self-correction, or retrieval-augmented verification pipelines.<n>We find that the conditional miscorrection rate is consistently dominant.<n>We characterizes detection threshold as a stable model/protocol-level regularity.
  arXiv  Detail & Related papers  (2026-05-26T18:26:15Z)
• ExComm: Exploration-Stage Communication for Error-Resilient Agentic Test-Time Scaling [57.42714978834704]
  ExComm is a communication protocol for exploration-stage agentic test-time scaling.<n>We show that ExComm consistently outperforms strong test-time scaling baselines.
  arXiv  Detail & Related papers  (2026-05-21T07:38:44Z)
• CyberCorrect: A Cybernetic Framework for Closed-Loop Self-Correction in Large Language Models [9.281774217584289]
  CyberCorrect is a framework that formalizes large language model self-correction.<n>A type-directed Correction Controller generates repair instructions based on diagnosed error categories.<n>A Convergence Judge determines iteration termination using stability criteria adapted from control theory.
  arXiv  Detail & Related papers  (2026-05-17T07:47:34Z)
• Seirênes: Adversarial Self-Play with Evolving Distractions for LLM Reasoning [56.48520300004217]
  We present Seirnes, a self-play RL framework that transforms contextual interference into an internal training signal.<n>A single model is trained to both construct plausible yet distracting contexts that expose its own reasoning blind spots.<n>By pitting these competing objectives against each other, Seirnes compels the model to move beyond superficial pattern matching.
  arXiv  Detail & Related papers  (2026-05-12T06:58:35Z)
• DiagnosticIQ: A Benchmark for LLM-Based Industrial Maintenance Action Recommendation from Symbolic Rules [4.124344125532972]
  Monitoring complex industrial assets relies on engineer-authored symbolic rules that trigger based on sensor conditions.<n> translating rules into maintenance steps requires asset-specific knowledge gained through years of practice.<n>We investigate whether LLMs can serve as decision support for this rule-to-action step and introduce ours, a benchmark of 6,690 expert-validated multiple-choice questions.
  arXiv  Detail & Related papers  (2026-05-09T02:17:39Z)
• Silent Commitment Failure in Instruction-Tuned Language Models: Evidence of Governability Divergence Across Architectures [0.0]
  We introduce governability -- the degree to which a model's errors are detectable before output commitment and correctable once detected.<n>We show benchmark accuracy does not predict governability, correction capacity varies independently of detection, and identical governance scaffolds produce opposite effects across models.<n>We propose a Detection and Correction Matrix classifying model-task combinations into four regimes: Governable, Monitor Only, Steer Blind, and Ungovernable.
  arXiv  Detail & Related papers  (2026-03-22T21:50:28Z)
• Decoding the Critique Mechanism in Large Reasoning Models [50.821607345799386]
  Large Reasoning Models (LRMs) exhibit backtracking and self-verification mechanisms that enable them to revise intermediate steps and reach correct solutions.<n>We investigate how current LRMs recover from errors by inserting arithmetic mistakes in their intermediate reasoning steps.<n>We find that despite the error propagating through the chain-of-thought, the model still reaches the correct final answer.
  arXiv  Detail & Related papers  (2026-03-17T10:03:30Z)
• CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal [84.71254539482369]
  Group-relative reinforcement learning with verifiable rewards (RLVR) often wastes the most informative data it already has the failures.<n>We present CARE, a failure-centric post-training framework for multimodal reasoning that turns errors into supervision.<n> CARE improves accuracy and training smoothness while explicitly increasing the share of learning signal that comes from failures.
  arXiv  Detail & Related papers  (2025-12-22T16:34:21Z)
• Metacognitive Self-Correction for Multi-Agent System via Prototype-Guided Next-Execution Reconstruction [58.51530390018909]
  Large Language Model based multi-agent systems excel at collaborative problem solving but remain brittle to cascading errors.<n>We present MASC, a metacognitive framework that endows MAS with real-time, unsupervised, step-level error detection and self-correction.
  arXiv  Detail & Related papers  (2025-10-16T05:35:37Z)
• Inverse Scaling in Test-Time Compute [51.16323216811257]
  Extending the reasoning length of Large Reasoning Models (LRMs) deteriorates performance.<n>We identify five distinct failure modes when models reason for longer.<n>These findings suggest that while test-time compute scaling remains promising for improving model capabilities, it may inadvertently reinforce problematic reasoning patterns.
  arXiv  Detail & Related papers  (2025-07-19T00:06:13Z)

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.