# VISTA: Auditing Semantic Divergence in Vision-Language Models

arXiv: https://arxiv.org/abs/2607.02995 · HTML: https://mlsanity.com/paper/2607.02995/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in VLM benchmarks and visual reasoning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 24 of its 31 references that are in the map.
- First posted on arXiv 3 Jul 2026.

## Placed in

- Multimodal LLMs / vision-language models / [VLM benchmarks and visual reasoning](https://mlsanity.com/mllm/vlm-benchmarks-and-visual-reasoning/)

## Builds on (most cited first, 24 of 24)

- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis](https://mlsanity.com/paper/2307.01952/) (2023)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks](https://mlsanity.com/paper/2312.14238/) (2023)
- [Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation](https://mlsanity.com/paper/2302.09664/) (2023)
- [LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models](https://mlsanity.com/paper/2407.07895/) (2024)
- [STRIP: A Defence Against Trojan Attacks on Deep Neural Networks](https://mlsanity.com/paper/1902.06531/) (2019)
- [FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age](https://mlsanity.com/paper/1908.04913/) (2019)
- [VL-Trojan: Multimodal Instruction Backdoor Attacks against Autoregressive Visual Language Models](https://mlsanity.com/paper/2402.13851/) (2024)
- [CleanCLIP: Mitigating Data Poisoning Attacks in Multimodal Contrastive Learning](https://mlsanity.com/paper/2303.03323/) (2023)
- [Phi-3 Safety Post-Training: Aligning Language Models with a "Break-Fix" Cycle](https://mlsanity.com/paper/2407.13833/) (2024)
- [Uncertainty-Aware Evaluation for Vision-Language Models](https://mlsanity.com/paper/2402.14418/) (2024)
- [LLaVA-MORE: A Comparative Study of LLMs and Visual Backbones for Enhanced Visual Instruction Tuning](https://mlsanity.com/paper/2503.15621/) (2025)
- [Unveiling the Tapestry of Consistency in Large Vision-Language Models](https://mlsanity.com/paper/2405.14156/) (2024)
- [Backdoor Attack on Vision Language Models with Stealthy Semantic Manipulation](https://mlsanity.com/paper/2506.07214/) (2025)
- [TIJO: Trigger Inversion with Joint Optimization for Defending Multimodal Backdoored Models](https://mlsanity.com/paper/2308.03906/) (2023)
- [MM-R3: On (In-)Consistency of Vision-Language Models (VLMs)](https://mlsanity.com/paper/2410.04778/) (2024)
- [ViLU: Learning Vision-Language Uncertainties for Failure Prediction](https://mlsanity.com/paper/2507.07620/) (2025)
- [Bridging the Gap in Vision Language Models in Identifying Unsafe Concepts Across Modalities](https://mlsanity.com/paper/2507.11155/) (2025)
- [Propaganda AI: An Analysis of Semantic Divergence in Large Language Models](https://mlsanity.com/paper/2504.12344/) (2025)
- [Improving the Sensitivity of Backdoor Detectors via Class Subspace Orthogonalization](https://mlsanity.com/paper/2512.08129/) (2025)
- [Intra-Class Probabilistic Embeddings for Uncertainty Estimation in Vision-Language Models](https://mlsanity.com/paper/2511.22019/) (2025)
- [Concept-Guided Backdoor Attack on Vision Language Models](https://mlsanity.com/paper/2512.00713/) (2025)

## Abstract (arXiv)

Vision-language models can exhibit visual concept-conditioned divergence: given images containing demographic features, corporate logos, or ideological symbols, some models produce unusually uniform responses that differ from what peer models say about the same input. These behaviors evade text-only audits because visual concepts cannot be isolated or substituted the way text tokens can. We present VISTA (Visual Inconsistency Screening Through Analysis), a black-box cross-model audit that couples semantic entropy with distribution-based divergence to flag model-specific anomalies. In a controlled study, we implant concept-conditioned stances in three VLMs via fine-tuning on small biased datasets and confirm that VISTA detects them. Auditing six VLMs across 19 topics, VISTA surfaces 142 high-suspicion cases (1.2%) and identifies selective refusal as a previously unreported divergence pattern, where models refuse demographic queries at rates varying from 0 to 65% across groups.