# HalluPrism: When Multimodal Uncertainty Should Diagnose, Not Decide

arXiv: https://arxiv.org/abs/2608.29193 · HTML: https://mlsanity.com/paper/2608.29193/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in MLLM training and visual instruction tuning, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 11 of its 42 references that are in the map.
- First posted on arXiv 29 Aug 2026.

## Placed in

- Multimodal LLMs / vision-language models / [MLLM training and visual instruction tuning](https://mlsanity.com/mllm/mllm-training-and-visual-instruction-tuning/)

## Builds on (most cited first, 11 of 11)

- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Selective Classification for Deep Neural Networks](https://mlsanity.com/paper/1705.08500/) (2017)
- [Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation](https://mlsanity.com/paper/2302.09664/) (2023)
- [Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding](https://mlsanity.com/paper/2311.16922/) (2023)
- [AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation](https://mlsanity.com/paper/2311.07397/) (2023)
- [Cross-Modal Attention Calibration for LVLM Hallucination Mitigation](https://mlsanity.com/paper/2501.01926/) (2025)
- [Mitigating Hallucination in Large Vision-Language Models via Adaptive Attention Calibration](https://mlsanity.com/paper/2505.21472/) (2025)
- [On Epistemic Uncertainty of Visual Tokens for Object Hallucinations in Large Vision-Language Models](https://mlsanity.com/paper/2510.09008/) (2025)

## Abstract (arXiv)

Multimodal Large Language Models (MLLMs) can assign similar confidence to answers that fail for different reasons. We propose HalluPrism, a behavioral diagnostic that re-runs an answer after visual degradation, blank-image replacement, and grounding or relation checks. These targeted probes yield a signature over visual-perturbation sensitivity (V ), image-removal confidence retention (L), and grounding/relation-probe instability (A). Across 58K+ examples from four benchmarks and four MLLMs, image-removal confidence retention is most prevalent, while grounding/relation-probe instability better separates failure families. Only 18 of 48 source-target checks are diagonally aligned, so the coordinates should be interpreted jointly rather than as independent causal sources. With the dataset fixed, the joint signature improves failure-family AUROC from 0.634 to 0.769 on HallusionBench and from 0.707 to 0.817 on VizWiz, with smaller gains on POPE and VSR. In pooled XGBoost analysis, AUROC rises from 0.78 with scalar confidence to 0.95 with (V, L, A) and 0.97 when confidence is added. The same signature does not automatically improve correctness ranking. The three tested direct scalarizations can harm it. These results separate failure diagnosis from abstention scoring: multimodal uncertainty should characterize failure structure before it is used to decide whether to abstain or correct.