How China-Origin Vision-Language Models Move from Refusal to Reframing in State Alignment

• URL: http://arxiv.org/abs/2608.11816v1
• Date: Wed, 12 Aug 2026 08:58:37 GMT
• Title: How China-Origin Vision-Language Models Move from Refusal to Reframing in State Alignment
• Abstract summary: State-aligned distortion has been documented in China-origin text-based large language models (LLMs)<n>We run nine vision-language models (VLMs), seven China-origin and two non-China, across four elicitation paradigms and two prompt languages, yielding 21,708 trials.<n>Each response is audited on six dimensions -- explicit refusal, information integrity, visual grounding, state-aligned framing, language consistency, and response length.
• Score: 8.359235542667806
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: State-aligned distortion has been documented in China-origin text-based large language models (LLMs), but whether, and in what form, it arises in multimodal systems has not been systematically examined. We construct a balanced benchmark of 200 core entries spanning ten politically sensitive topics, plus a seven-variant visual-abstraction probe, and run nine vision-language models (VLMs), seven China-origin and two non-China, across four elicitation paradigms and two prompt languages, yielding 21,708 trials. Each response is audited on six dimensions -- explicit refusal, information integrity, visual grounding, state-aligned framing, language consistency, and response length -- by two independent frontier LLM judges, validated against three human experts on a 200-trial sample. Measuring each dimension separately lets us decompose multimodal censorship into individual signals rather than a single refusal-based score; in particular, refusal and framing are measured independently, so a model can stop refusing while still reframing. We find that (i) Chinese-language prompting roughly triples the odds of state-aligned framing, within every model; (ii) China-origin models reframe more than non-China models (direction robust across judges and human raters; magnitude 1.6--3.2x); (iii) the effect is strongest in text-only political commentary (36.5%) and is gated by recognition of the depicted subject rather than pixel detail, persisting even at silhouette for iconic images; and (iv) across four Qwen generations, state-aligned framing rises while explicit refusal falls: censorship migrates from a visible act (refusal) to an invisible one (fluent reframing). We argue this shift to invisible reframing is fundamentally a problem of human-AI interaction: it removes the very signal users rely on to recognize that information has been withheld.

Related papers

• When Seeing Overrides Knowing: Visual Dominance and Deferral-Based Method for Personalized Safety in VLMs [54.95635713318535]
  Vision-language models (VLMs) are increasingly deployed in high-stakes settings, where a response that is reasonable in general may still be unsafe for a particular user.<n>We study this problem of personalized safety in multimodal systems and introduce MPS-Bench, a benchmark of 5,181 scenarios from 584 real-world images.<n>We find that they almost always respond directly (86-99%) rather than seek missing context, and none exceeds 2.6/5 on personalized safety.<n>Motivated by this mechanism, we propose PRISM, a lightweight input monitor that uses bidirectional cross-modal modulation to predict when a query is likely to require
  arXiv  Detail & Related papers  (2026-09-03T03:58:56Z)
• Said Aloud, Read Different: Cross-Modal Instability in Multimodal Models [17.521958873911846]
  We introduce a speech-augmented contrastive triplet benchmark spanning 10,150 culturally grounded images from 18 MENA countries.<n>We find that modality and language shifts introduce substantial triplet-level inconsistencies that are not fully captured by aggregate accuracy.
  arXiv  Detail & Related papers  (2026-08-27T13:46:49Z)
• Knowing When Not to Answer: Evaluating Abstention in Multimodal Reasoning Systems [52.83669998269706]
  Abstention has been studied in text-only settings but remains underexplored multimodally.<n>Current benchmarks either ignore unanswerability or rely on coarse methods that miss realistic failure modes.<n>We introduce MM-AQA, a benchmark that constructs unanswerable instances from answerable ones via transformations along two axes.
  arXiv  Detail & Related papers  (2026-04-16T09:23:22Z)
• Internal Knowledge Without External Expression: Probing the Generalization Boundary of a Classical Chinese Language Model [2.096023520617773]
  We train a language model from scratch on a curated corpus of 1.56 billion tokens of pure Classical Chinese.<n>We investigate whether the model can distinguish known from unknown inputs.<n>We find a clear dissociation between internal and external uncertainty.
  arXiv  Detail & Related papers  (2026-03-31T07:37:15Z)
• Bilingual Bias in Large Language Models: A Taiwan Sovereignty Benchmark Study [0.0]
  Large Language Models (LLMs) are increasingly deployed in multilingual contexts, yet their consistency across languages on politically sensitive topics remains understudied.<n>This paper presents a systematic benchmark study examining how 17 LLMs respond to questions concerning the sovereignty of the Republic of China (Taiwan) when queried in Chinese versus English.<n>We discover significant language bias -- the phenomenon where the same model produces substantively different political stances depending on the query language.
  arXiv  Detail & Related papers  (2026-02-06T03:57:21Z)
• When Alignment Fails: Multimodal Adversarial Attacks on Vision-Language-Action Models [75.16145284285456]
  We introduce VLA-Fool, a comprehensive study of multimodal adversarial robustness in embodied VLA models under both white-box and black-box settings.<n>We develop the first automatically crafted and semantically guided prompting framework.<n> Experiments on the LIBERO benchmark reveal that even minor multimodal perturbations can cause significant behavioral deviations.
  arXiv  Detail & Related papers  (2025-11-20T10:14:32Z)
• LIBERO-Plus: In-depth Robustness Analysis of Vision-Language-Action Models [49.92148175114169]
  We perform a systematic vulnerability analysis by introducing controlled perturbations across seven dimensions.<n>Models exhibit extreme sensitivity to perturbation factors, including camera viewpoints and robot initial states.<n>Surprisingly, models are largely insensitive to language variations, with further experiments revealing that models tend to ignore language instructions completely.
  arXiv  Detail & Related papers  (2025-10-15T14:51:36Z)
• Source framing triggers systematic evaluation bias in Large Language Models [0.0]
  This study systematically examines inter- and intra-model agreement across four state-of-the-art Large Language Models (LLMs)<n>We find that, in the blind condition, different LLMs display a remarkably high degree of inter- and intra-model agreement across topics.<n>Our findings reveal that framing effects can deeply affect text evaluation, with significant implications for the integrity, neutrality, and fairness of LLM-mediated information systems.
  arXiv  Detail & Related papers  (2025-05-14T07:42:27Z)
• Do Vision-Language Models Represent Space and How? Evaluating Spatial Frame of Reference Under Ambiguities [27.940469021840745]
  We present an evaluation protocol to assess the spatial reasoning capabilities of vision-language models (VLMs)<n>Despite some alignment with English conventions in resolving ambiguities, our experiments reveal significant shortcomings of VLMs.<n>With a growing effort to align vision-language models with human cognitive intuitions, we call for more attention to the ambiguous nature and cross-cultural diversity of spatial reasoning.
  arXiv  Detail & Related papers  (2024-10-22T19:39:15Z)
• Asymmetric Modality Translation For Face Presentation Attack Detection [55.09300842243827]
  Face presentation attack detection (PAD) is an essential measure to protect face recognition systems from being spoofed by malicious users. We propose a novel framework based on asymmetric modality translation forPAD in bi-modality scenarios. Our method achieves state-of-the-art performance under different evaluation protocols.
  arXiv  Detail & Related papers  (2021-10-18T08:59:09Z)
• Read Like Humans: Autonomous, Bidirectional and Iterative Language Modeling for Scene Text Recognition [80.446770909975]
  Linguistic knowledge is of great benefit to scene text recognition. How to effectively model linguistic rules in end-to-end deep networks remains a research challenge. We propose an autonomous, bidirectional and iterative ABINet for scene text recognition.
  arXiv  Detail & Related papers  (2021-03-11T06:47:45Z)

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.