Role-Break in Attention Heads: Understanding and Detecting Hallucinations in VLMs

• URL: http://arxiv.org/abs/2607.29412v1
• Date: Fri, 31 Jul 2026 13:33:51 GMT
• Title: Role-Break in Attention Heads: Understanding and Detecting Hallucinations in VLMs
• Abstract summary: Vision-Language Models (VLMs) remain prone to hallucinations, producing content that is inconsistent with or unsupported by the input image.<n>Under a unified head-level view, we find that hallucination-induced changes manifest as localized deviations from each head's faithful contextual behavior.<n> Detailed analysis reveals that these deviations are systematically organized across attention heads, contextual sources, and deviation directions, and that the resulting signal is linearly readable once head identity is preserved.
• Score: 49.743008490747094
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Despite remarkable progress in vision-language generation, Vision-Language Models (VLMs) remain prone to hallucinations, producing content that is inconsistent with or unsupported by the input image. Existing works largely design detection or mitigation methods around one specific hallucination pattern, such as visual-textual imbalance, but real VLM hallucinations arise from a mixture of multiple patterns, so signals bound to a single pattern struggle to remain stable across models and tasks. Under a unified head-level view, we find that hallucination-induced changes manifest as localized deviations from each head's faithful contextual behavior, a phenomenon we term Role-Break. Detailed analysis reveals that these deviations are systematically organized across attention heads, contextual sources, and deviation directions, and that the resulting signal is linearly readable once head identity is preserved. Based on these findings, we build a lightweight linear detector on top of Role-Break that requires no fine-tuning of the VLM, whose feature dimension stays below 5,000 and reaches an average AUROC of 93.23 across six VLMs and four benchmarks. A small-scale intervention experiment further shows that the detected tokens can be directly acted upon in the discriminative setting.

Related papers

• Diagnosing Visual Ignorance in Vision-Language Models [29.2851901986069]
  Vision-Language Models (VLMs) frequently rely on language priors, producing confident answers that are weakly grounded in visual evidence.<n>In this work, we study language-prior reliance from both mechanistic and behavioral perspectives.<n>We find that language-prior reliance is a systematic routing failure affecting both model internals and benchmark validity.
  arXiv  Detail & Related papers  (2026-06-05T04:16:01Z)
• Mitigating Entangled Steering in Large Vision-Language Models for Hallucination Reduction [49.96701537295129]
  Large Vision-Language Models (LVLMs) have achieved remarkable success across cross-modal tasks but remain hindered by hallucinations.<n>Existing methods mitigate hallucinations but often alter generation behavior, resulting in shorter outputs and shifted token distributions.<n>We propose MESA, an effective plug-and-play framework that performs controlled and selective latent intervention for hallucination mitigation.
  arXiv  Detail & Related papers  (2026-04-09T07:31:27Z)
• Hallucination Begins Where Saliency Drops [18.189047289404325]
  hallucinations frequently arise when preceding output tokens exhibit low saliency toward the prediction of the next token.<n>We introduce LVLMs-Saliency, a gradient-aware diagnostic framework that quantifies the visual grounding strength of each output token.<n>Our method significantly reduces hallucination rates while preserving fluency and task performance, offering a robust and interpretable solution.
  arXiv  Detail & Related papers  (2026-01-28T05:50:52Z)
• VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck [43.74340618057423]
  We propose VIB-Probe, a novel hallucination detection and mitigation framework.<n>Our method extracts discriminative patterns across layers and heads while filtering out semantic nuisances.<n>Our code will be made publicly available.
  arXiv  Detail & Related papers  (2026-01-09T05:58:22Z)
• Mechanisms of Prompt-Induced Hallucination in Vision-Language Models [58.991412160253276]
  We study the failure mode in a controlled object-counting setting, where the prompt overstates the number of objects in the image.<n>We identify a small set of attention heads whose ablation substantially reduces prompt-induced hallucinations (PIH) by at least 40% without additional training.<n>Our findings offer insights into the internal mechanisms driving prompt-induced hallucinations, revealing model-specific differences in how these behaviors are implemented.
  arXiv  Detail & Related papers  (2026-01-08T18:23:03Z)
• Revealing Perception and Generation Dynamics in LVLMs: Mitigating Hallucinations via Validated Dominance Correction [59.801614364841775]
  Large Vision-Language Models (LVLMs) have shown remarkable capabilities, yet hallucinations remain a persistent challenge.<n>This work presents a systematic analysis of the internal evolution of visual perception and token generation in LVLMs.<n>We devise the VDC (d Dominance Correction) strategy, which detects unsupported tokens and replaces them with validated ones to improve output reliability.
  arXiv  Detail & Related papers  (2025-12-21T17:05:42Z)
• SHALE: A Scalable Benchmark for Fine-grained Hallucination Evaluation in LVLMs [52.03164192840023]
  Large Vision-Language Models (LVLMs) still suffer from hallucinations, i.e., generating content inconsistent with input or established world knowledge.<n>We propose an automated data construction pipeline that produces scalable, controllable, and diverse evaluation data.<n>We construct SHALE, a benchmark designed to assess both faithfulness and factuality hallucinations.
  arXiv  Detail & Related papers  (2025-08-13T07:58:01Z)
• Cracking the Code of Hallucination in LVLMs with Vision-aware Head Divergence [69.86946427928511]
  We investigate the internal mechanisms driving hallucination in large vision-language models (LVLMs)<n>We introduce Vision-aware Head Divergence (VHD), a metric that quantifies the sensitivity of attention head outputs to visual context.<n>We propose Vision-aware Head Reinforcement (VHR), a training-free approach to mitigate hallucination by enhancing the role of vision-aware attention heads.
  arXiv  Detail & Related papers  (2024-12-18T15:29:30Z)
• Reducing Hallucinations in Vision-Language Models via Latent Space Steering [34.1755878632361]
  Hallucination poses a challenge to the deployment of large vision-language models (LVLMs) in applications. We introduce Visual and Textual Intervention (VTI), a novel technique designed to reduce hallucinations by steering latent space representations during inference to enhance the stability of vision features.
  arXiv  Detail & Related papers  (2024-10-21T08:42:30Z)

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.