Abstention in VLMs

Vision-language models answer questions about images they were never shown, a failure called mirage. Such an answer can be correct without being justified by the evidence, and the difference matters wherever a model's output is checked by someone who did not see its input. We locate the failure in the training objective: cross-entropy separates a model that reads the image from one that ignores it by only the conditional mutual information between answer and image, and no training example rewards declining. We formalise support completion, TIDE, which also shows each training question with a blank image, a same-domain wrong image and an out-of-domain image, with targets that decline, and prove that under TIDE every image-blind predictor is suboptimal by at least 1.04 nats per example, that the optimum declines exactly on the counterfactual kinds while keeping its behaviour on valid images, and that for any finite model the error of its answer-or-decline decision, declining on valid input plus answering on invalid input, is bounded below by how well its representation separates valid from invalid evidence. On Qwen2.5-VL-7B under a frozen held-out protocol with two judges, the three-condition predecessor of TIDE reduced the invalid-evidence mirage rate from 0.759 to 0.097 and the blank-image rate from 0.700 to 0.003 at a false-abstention cost of 0.126; label-free grounding raised the rate to 0.859, a likelihood-advantage contrast moved it only to 0.718 because its empty-image term carries no gradient, and grounding applied after support completion re-opened the mirage. The four-condition protocol on three base models, with seeds and component ablations, is in progress and not reported here.

As vision language models are increasingly deployed in clinical diagnosis, under standing how they internally resolve competing visual and textual signals becomes a safety imperative. Existing mechanistic analyses remain confined to unimodal text and offer no explanation for why a single misleading sentence can override a correct image based diagnosis, or why a model commits to a confident answer despite insufficient visual evidence. We find that these two safety risks, arbitra tion failure where textual context overrides visual grounding and brake failure where the model commits without adequate evidence, are mediated by spatially disjoint attention head populations: arbitration heads form a mid-to-deep wideband reflecting cross-layer evidence competition, while brake heads concentrate in a narrow middle-to-late layer band that regulates evidence sufficiency and abstention behavior. To ground these observations in causal circuitry, we introduce CRAFT, which localizes each failure mode to a minimal causal head set via dual criteria and verifies necessity and sufficiency through temporal probes and Tuned Lens trajectory analysis. Excising arbitration heads sharply reduces conflict following with negligible degradation on clean inputs, while excising brake heads restores ap propriate abstention under degraded visual evidence. The two interventions target spatially disjoint head sets and produce distinct corrective effects, underscoring the mechanistic separability of the failure modes. Experiments across multiple medical VQA benchmarks and VLM architectures validate both the localization and inter ventions, demonstrating that the identified heads causally drive each failure mode and that targeted modulation generalises without retraining. The code is available at https://github.com/zhcz328/CRAFT.

The ability of vision-language models (VLMs) to abstain from unanswerable questions is as important as their ability to answer answerable ones accurately. Recently, several benchmarks have emerged to evaluate and improve VLM abstention, but they have substantial limitations. First, samples often contain shortcut cues in images or questions that reveal answerability, while an explicit "unanswerable" option further prevents accurate assessment of spontaneous abstention. Second, as training data, they generally provide only binary labels without fine-grained explanations for deeper supervision. To address these limitations, we introduce Visual Answerability Diagnosis with Rationales (VAD-R), a benchmark constructed through a two-stage pipeline of shortcut filtering and quality verification to prevent answerability leakage. Each example is annotated with step-by-step rationales and causal evidence-gap labels. Evaluation of state-of-the-art open- and closed-source VLMs on VAD-R reveals limited spontaneous abstention, with average recall rates of only 11.4% and 16.3%, respectively. Probing analyses show that hidden-state representations in certain layers can effectively distinguish answerability, yet this distinction fails to manifest in final responses. Motivated by this observation, we introduce Rep2Act, a representation-to-action alignment method that translates latent answerability awareness into explicit abstention decisions. Rep2Act improves action accuracy on VAD-R from 56.67% to 86.33% for Qwen2.5-VL-3B and from 59.33% to 88.67% for Qwen2.5-VL-7B. On the out-of-distribution TUBench, Rep2Act achieves an average F1 score of 53.3% with only a 3B model, surpassing the closed-source GPT-4 Turbo and GPT-4o by 16.2% and 1.1%, respectively.

Many medical conditions require clinical assessment of symptoms, history, and functioning rather than judgments from appearance. We study whether vision-language models (VLMs) abstain from facial attribution of autism spectrum disorder (ASD), attention-deficit/hyperactivity disorder (ADHD), major depressive disorder (MDD), and generalized anxiety disorder (GAD). To support this study, we construct PARITY (Paired Assessment with Reused Identity), a demographically balanced dataset for testing whether changing facial expressions alters diagnostic judgments, while holding identity and other visual factors as constant as possible. Our comprehensive evaluation of 27 open- and closed-weight models reveals inconsistent abstention across expressions, conditions, and task formats. Models turn facial expressions into unsupported medical judgments: this behavior is widespread among open-weight VLMs and extends to safety-trained frontier models that abstain on individual portraits yet make diagnostic judgments when comparing portraits of the same identity. Safeguard checks reveal further gaps: clinical instructions do not consistently prevent these judgments, and a guardrail model rarely flags paired requests as unsafe. Semantic analysis further shows that most baseline open-model refusals fail to explain both the insufficiency of facial evidence and the need for clinical assessment. Unsupported judgments and seemingly authoritative explanations can cause distress, reinforce stigma, and offer false reassurance, underscoring the need for stronger safety measures in model training and deployment.

Aligned vision-language models (VLMs) are designed to balance grounded visual reasoning with safe generation behavior. However, we observe a striking phenomenon: under safety-constrained instruction, models frequently abstain from answering questions that remain correctly answerable under default instruction despite receiving identical image-question inputs. This raises a fundamental question: does safety alignment suppress perceptual grounding itself, or does visual evidence remain internally available while generation is redirected toward abstention? In this work, we investigate the internal decoding dynamics underlying safety-induced abstention in aligned VLMs. Across multiple architectures and multimodal benchmarks, we show that abstained generations remain consistently influenced by visual evidence throughout decoding, indicating that perceptual grounding is largely preserved despite refusal behavior. We further demonstrate that, although the representational organization of refusal differs substantially across architectures, safety-constrained instruction consistently alters late-stage hidden-state dynamics toward refusal-oriented decoding. Finally, through targeted activation-level interventions, we show that suppressing refusal-related representations reliably restores grounded answering behavior across models without retraining or modifying visual inputs. Together, these findings reveal a previously underexplored failure mode in aligned VLMs: safety alignment can override grounded visual expression even when perceptual evidence remains internally preserved.

When visual evidence is occluded or chaotic, models should abstain. In this paper, we show that Vision-Language Models (VLMs) can internally distinguish when abstention is required, but fail to express it anyway. We introduce TRAPSBench, a procedurally generated video benchmark of 1,404 matched physics pairs in which a single targeted change renders the outcome undeterminable from the visual evidence. Furthermore, we introduce Penalized Epistemic Calibration Score (PECS), a new robust metric that requires models to both answer correctly when the outcome is knowable, and abstain when the outcome is not. Across 16 VLMs spanning five families, spontaneous restraint is poor: the best PECS is 0.292. The bottleneck is expression, not perception: linear probes decode answerability from hidden states at up to 0.91 AUROC across physics domains; steering a single-layer void direction causally induces or suppresses abstention. Our results replicate across three open-weight families (Qwen, Gemma, LLaVA). The failure is also more pronounced in visual than textual uncertainty: models detect textual impossibility about 4x more readily than missing visual evidence. Closing this representation--output gap likely requires output-stage interventions.

Large vision-language models (LVLMs) hallucinate: they assert visual details that the image does not support. A principled remedy is selective prediction with a distribution-free guarantee-verify each claim and abstain when the claim is not grounded, so that the hallucination rate among asserted claims is provably bounded. We show, however, that this guarantee is bought at a brutal price: to keep the hallucination rate below $5\%$ on a balanced object-existence benchmark, a state-of-the-art conformal filter must abstain on more than $80\%$ of claims. We argue that abstention is wasteful when more visual evidence is cheaply available, and introduce Budgeted Conformal Evidence Acquisition (BCEA), which replaces the binary answer/abstain decision with a three-way choice: answer, abstain, or acquire additional visual evidence by re-examining the image (zooming, cropping, or applying a claim-specific intervention) under a bounded compute budget. We make two observations. First, acquisition that is plugged naively into a calibrated filter breaks the statistical guarantee -- realized risk overshoots the target by up to $17$ points -- because the acquisition step destroys the exchangeability that conformal calibration relies on. Second, folding the entire acquisition policy into the score function and re-calibrating on post-acquisition scores restores the finite-sample guarantee while still recovering coverage. BCEA further uses structured, claim-type-specific interventions. Across the POPE benchmark and COCO-constructed existence and spatial-relation claims, on four open VLMs, BCEA controls the hallucination rate at the target level and consistently improves coverage over a guaranteed-abstention baseline.

Vision-language models (VLMs) are increasingly used to generate structured descriptions of street-level imagery for tasks such as streetscape auditing, mapping, and public consultation. These uses combine observable attributes with appraisal categories, and the human targets are often distributions of judgments with disagreement and explicit non-response. This paper argues that benchmarking VLMs for urban perception should treat disagreement and abstention as measurement outcomes, report inter-annotator reliability alongside model alignment, and treat the label space and scoring policy as negotiable artifacts when outputs are intended to inform urban governance. We ground the argument in a benchmark of 100 Montreal street scenes annotated along 30 dimensions by 12 participants from seven community organizations, and in a deterministic zero-shot evaluation of seven VLMs. Across dimensions, model agreement with human consensus co-varies with dimension-level human reliability, and for the appraisal dimension Overall Impression models and annotators exhibit distributional mismatch including different rates of Not applicable. We close with actions for benchmark creators, model developers, and institutions to make uncertainty and benchmark assumptions visible in evaluation reports.

Large Vision-Language Models (VLMs) have achieved remarkable multimodal performance yet remain prone to factual hallucinations, particularly in long-tail or specialized domains. Moreover, current models exhibit a weak capacity to refuse queries that exceed their parametric knowledge. In this paper, we propose a systematic framework to enhance the refusal capability of VLMs when facing such unknown questions. We first curate a model-specific "Visual-Idk" (Visual-I don't know) dataset, leveraging multi-sample consistency probing to distinguish between known and unknown facts. We then align the model using supervised fine-tuning followed by preference-aware optimization (e.g., DPO, ORPO) to effectively delineate its knowledge boundaries. Results on the Visual-Idk dataset show our method improves the Truthful Rate from 57.9% to 67.3%. Additionally, internal probing also demonstrates that the model genuinely recognizes its boundaries instead of just memorizing refusal patterns. Our framework further generalizes to out-of-distribution medical and perceptual domains, providing a robust path toward more trustworthy and prudent visual assistants.

Recent Vision-Language Models (VLMs) have made remarkable progress in multimodal understanding tasks, yet their evaluation on long video understanding remains unreliable. Due to limited frame inputs, key frames necessary for answering the question may be missing from the model's input. However, models that truthfully refuse to answer under such uncertainty are marked as incorrect, while those that guess may coincidentally produce the correct answer and thus obtain deceptively higher accuracy, leading to misleading evaluation results and encouraging models to guess rather than respond honestly. To address this issue, we introduce VirtueBench, a benchmark explicitly designed to assess model trustworthiness under uncertainty. VirtueBench constructs multiple frame-sampling levels for each video and provides ground truths that distinguish between answerable and unanswerable cases. Evaluations on 25 open-source and commercial VLMs reveal distinct refusal behaviors across different model families, with refusal accuracy ranging from over 70% in the best models to nearly 0% in the worst. Moreover, most models exhibit a substantial drop in refusal when the prompt does not explicitly require them to do so. These findings highlight the need for developing trustworthy VLMs for multimodal understanding, guided by benchmarks and leaderboards that emphasize reliability and trustworthiness.