# HALP: Detecting Hallucinations in Vision-Language Models without Generating a Single Token

arXiv: https://arxiv.org/abs/2603.05465 · HTML: https://mlsanity.com/paper/2603.05465/

## Key facts

- Cited by 14 papers in the map (data through 14 Oct 2026).
- Placed in Hallucination mitigation in LVLMs, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 19 of its 32 references that are in the map.
- First posted on arXiv 5 Mar 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Hallucination mitigation in LVLMs](https://mlsanity.com/mllm/hallucination-mitigation-in-lvlms/)

## Builds on (most cited first, 19 of 19)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Language Models (Mostly) Know What They Know](https://mlsanity.com/paper/2207.05221/) (2022)
- [MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts](https://mlsanity.com/paper/2310.02255/) (2023)
- [AdaptFormer: Adapting Vision Transformers for Scalable Visual Recognition](https://mlsanity.com/paper/2205.13535/) (2022)
- [Designing and Interpreting Probes with Control Tasks](https://mlsanity.com/paper/1909.03368/) (2019)
- [AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluation](https://mlsanity.com/paper/2311.07397/) (2023)
- [LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations](https://mlsanity.com/paper/2410.02707/) (2024)
- [Targeted Syntactic Evaluation of Language Models](https://mlsanity.com/paper/1808.09031/) (2018)
- [PaliGemma 2: A Family of Versatile VLMs for Transfer](https://mlsanity.com/paper/2412.03555/) (2024)
- [Object Detectors Emerge in Deep Scene CNNs](https://mlsanity.com/paper/1412.6856/) (2014)
- [HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding](https://mlsanity.com/paper/2403.00425/) (2024)
- [Fine-grained Hallucination Detection and Editing for Language Models](https://mlsanity.com/paper/2401.06855/) (2024)
- [Do Androids Know They're Only Dreaming of Electric Sheep?](https://mlsanity.com/paper/2312.17249/) (2023)
- [Estimating Knowledge in Large Language Models Without Generating a Single Token](https://mlsanity.com/paper/2406.12673/) (2024)
- [Visual Description Grounding Reduces Hallucinations and Boosts Reasoning in LVLMs](https://mlsanity.com/paper/2405.15683/) (2024)
- [HaloQuest: A Visual Hallucination Dataset for Advancing Multimodal Reasoning](https://mlsanity.com/paper/2407.15680/) (2024)
- [From Behavioral Performance to Internal Competence: Interpreting Vision-Language Models with VLM-Lens](https://mlsanity.com/paper/2510.02292/) (2025)
- [From Head to Tail: Towards Balanced Representation in Large Vision-Language Models through Adaptive Data Calibration](https://mlsanity.com/paper/2503.12821/) (2025)

## Built on by (most cited first, 14 of 14)

- [Dual-Pathway Circuits of Object Hallucination in Vision-Language Models](https://mlsanity.com/paper/2605.13156/) (2026)
- [Seeing without Looking: Do Vision-Language Benchmarks Really Test Vision?](https://mlsanity.com/paper/2605.22903/) (2026)
- [Pre-Generation Hallucination Detection in Large Language Models via Soft-Target Attention Probing](https://mlsanity.com/paper/2606.21917/) (2026)
- [Failing to See or Failing to Know? Attributing Errors in Vision-Language Models](https://mlsanity.com/paper/2607.04683/) (2026)
- [Hallucinations Leave a Grounding Signature:Verifier-Guided Decoding for Selective Object Correction](https://mlsanity.com/paper/2607.27823/) (2026)
- [CRAFT: Causal Responsibility and Failure Tracing in Medical Vision Language Models](https://mlsanity.com/paper/2609.38810/) (2026)
- [Evaluating Reasoning Fidelity in Visual Text Generation](https://mlsanity.com/paper/2606.04479/) (2026)
- [Witness Evidence Portfolios: Single-Prefill Risk Detection for Closed Multimodal Answers](https://mlsanity.com/paper/2607.27667/) (2026)
- [UniProbe: A Learnable Token-Level Hallucination Detector for Large VLMs using Multi-Structural Internal Representations](https://mlsanity.com/paper/2608.10835/) (2026)
- [PatchGate: Narrowing the Verbalization Gap with Intrinsic Object Inventories in Frozen Vision-Language Models](https://mlsanity.com/paper/2608.21819/) (2026)
- [Decodable but Misrouted: Sparse Features Uncover a Readout Gap in Vision-Language Models for Harmful Meme Detection](https://mlsanity.com/paper/2609.18860/) (2026)
- [RECAP: Relation Evidence Calibration for Detecting Spatial Relation Hallucinations in Vision-Language Models](https://mlsanity.com/paper/2609.26093/) (2026)
- [From Routing Signals to Selective Review: Visual regrounding in MoE VLMs](https://mlsanity.com/paper/2609.38111/) (2026)
- [DEPICT: Scoring Text-to-Image Alignment by Answer Agreement](https://mlsanity.com/paper/2610.03617/) (2026)

## Abstract (arXiv)

Hallucinations remain a persistent challenge for vision-language models (VLMs), which often describe nonexistent objects or fabricate facts. Existing detection methods typically operate after text generation, making intervention both costly and untimely. We investigate whether hallucination risk can instead be predicted before any token is generated by probing a model's internal representations in a single forward pass. Across a diverse set of vision-language tasks and eight modern VLMs, including Llama-3.2-Vision, Gemma-3, Phi-4-VL, and Qwen2.5-VL, we examine three families of internal representations: (i) visual-only features without multimodal fusion, (ii) vision-token representations within the text decoder, and (iii) query-token representations that integrate visual and textual information before generation. Probes trained on these representations achieve strong hallucination-detection performance without decoding, reaching up to 0.93 AUROC on Gemma-3-12B, Phi-4-VL 5.6B, and Molmo 7B. Late query-token states are the most predictive for most models, while visual or mid-layer features dominate in a few architectures (e.g., ~0.79 AUROC for Qwen2.5-VL-7B using visual-only features). These results demonstrate that (1) hallucination risk is detectable pre-generation, (2) the most informative layer and modality vary across architectures, and (3) lightweight probes have the potential to enable early abstention, selective routing, and adaptive decoding to improve both safety and efficiency.