New benchmarks and methods tackle LLM hallucinations across modalities and domains
ByPulseAugur Editorial·[53 sources]·
Researchers are developing new methods and benchmarks to detect and mitigate hallucinations in large language models (LLMs) across various modalities and domains. OmniHallu offers a unified framework for detecting hallucinations in multimodal comprehension and generation tasks, supported by a new benchmark. Another approach focuses on domain-specific hallucination detection, combining classification, uncertainty quantification, and preference optimization to reduce errors in models like Qwen2.5. For legal applications, LexAgentHallu is introduced as a benchmark to profile hallucinations in tool-augmented legal agents, highlighting agentic failures. Additionally, Evidence-Aligned Entity Verification (EAEV) uses retrieval-augmented generation to ground outputs and detect entity-level hallucinations, while TRACE offers a training-free method to correct hallucinations by analyzing cross-layer evidence within LLMs.
AI
IMPACT
Advances in hallucination detection are crucial for improving the reliability and trustworthiness of LLMs across diverse applications.
RANK_REASON
Multiple research papers introducing new benchmarks and methods for hallucination detection in LLMs.
arXiv:2609.20195v1 Announce Type: cross Abstract: Audio-language models increasingly generate confident music descriptions that are unsupported by the input audio. We present, to our knowledge, the first music-specific, layer-wise, multi-paradigm empirical study of hallucination …
arXiv cs.CL
TIER_1English(EN)·Shardul P. More, Tanuja S. Pawar·
arXiv:2609.18320v1 Announce Type: new Abstract: Large Language Models (LLMs) frequently exhibit hallucinations, presenting a major barrier to reliability in complex reasoning tasks. While traditional detection methods rely on output-based confidence metrics, these logits are ofte…
arXiv:2602.02888v2 Announce Type: replace-cross Abstract: Hallucinations remain a major obstacle for large language models (LLMs), especially in safety-critical domains. We present HALT (Hallucination Assessment via Log-probs as Time series), a lightweight hallucination detector …
arXiv cs.AI
TIER_1English(EN)·Toqeer Ehsan, Nico Penttil\"a, Richard Schmidt, Arash Hajikhani, Victoria Palacin·
arXiv:2609.17327v1 Announce Type: cross Abstract: This paper describes our submission to the SHROOM-Visions shared task on detecting and classifying hallucinated character spans in vision-language model outputs across four languages. We employ several fine-tuned vision-language m…
arXiv:2609.14448v1 Announce Type: cross Abstract: Hallucination in large language models reduces their reliability and slows adoption. Various white-box studies have used internal representations to detect patterns of truthfulness and factuality. A less-studied approach is to ide…
arXiv:2609.15106v1 Announce Type: new Abstract: Large language models can hallucinate even when the knowledge required for a correct answer is already available. We study this failure through a latent-key view of inference, where answer selection depends on competition among asso…
arXiv:2410.15359v2 Announce Type: replace Abstract: Multimodal Foundation Models represent a significant leap in artificial intelligence. Among them, Large Vision-Language Models (LVLMs) serve as the typical representative of these foundation models, which integrate visual modali…
arXiv:2511.07318v3 Announce Type: replace Abstract: Despite substantial advances, large language models (LLMs) continue to exhibit hallucinations, generating plausible yet incorrect responses. In this paper, we highlight a critical yet previously underexplored class of hallucinat…
arXiv:2609.12111v1 Announce Type: new Abstract: Factual hallucination in closed-book question answering is often treated as a coverage problem: a model fails because the relevant fact is absent from its internal memory. This view misses a second source of error. Even when a fact …
arXiv:2609.11244v1 Announce Type: new Abstract: While Multimodal Large Language Models (MLLMs) have achieved remarkable progress across diverse tasks, they suffer from hallucinations where generated outputs contradict or misrepresent input semantics. Existing research typically a…
arXiv:2609.09754v1 Announce Type: new Abstract: As large language models are increasingly deployed as tool-augmented legal agents, they introduce agentic hallucinations where tool-call and reasoning errors cascade into fabricated holdings and miscited authority. However, existing…
arXiv:2609.11878v1 Announce Type: new Abstract: Large language models generate fluent text that can contain unfaithful claims -- a phenomenon known as hallucination. We present a multi-signal detection pipeline combining fine-tuned DeBERTa-v3 classification, Monte Carlo (MC) Drop…
Large language models generate fluent text that can contain unfaithful claims -- a phenomenon known as hallucination. We present a multi-signal detection pipeline combining fine-tuned DeBERTa-v3 classification, Monte Carlo (MC) Dropout uncertainty quantification, and temperature-…
arXiv cs.AI
TIER_1English(EN)·Runsong Jia, Zhen Fang, Mengjia Wu, Jie Lu, Yi Zhang·
arXiv:2609.08267v1 Announce Type: new Abstract: Hallucination detection is crucial for large language models (LLMs), as hallucinated content creates significant barriers in applications requiring factual accuracy. Current detection methods mainly depend on internal signals like u…
arXiv:2605.18163v2 Announce Type: replace Abstract: Hallucination correction is not a one-direction problem. We show that intermediate layers are neither uniformly more truthful than final layers nor uniformly less trustworthy. Yet hallucination reduction is usually instantiated …
arXiv cs.AI
TIER_1English(EN)·Daria Voronkova, Ilya Trofimov, Anton Dmitriev, Eduard Tulchinskii, Evgeny Burnaev, Serguei Barannikov·
arXiv:2609.07779v1 Announce Type: cross Abstract: As AI-code assistant tools become widespread, automatic assessment of the correctness of generated code becomes a significant challenge. Code LLMs are prone to hallucinations, which may lead to code that does not solve the require…
arXiv:2609.10244v1 Announce Type: new Abstract: We present our system for the SHROOM-Visions 2026 shared task on character-level VLM hallucination detection. A small ($4$B-parameter) VLM is fine-tuned as a per-token classifier reading a two-token feature from its own hidden state…
Video-language models and video agents can produce hallucinations that conflict with spatiotemporal evidence. Existing benchmarks mainly evaluate model hallucinations, and heterogeneous mechanisms make detector reliability difficult to compare. We introduce VidHalLoc, a benchmark…
As large language models are increasingly deployed as tool-augmented legal agents, they introduce agentic hallucinations where tool-call and reasoning errors cascade into fabricated holdings and miscited authority. However, existing legal benchmarks evaluate only single-turn QA w…
Hallucination detection is crucial for large language models (LLMs), as hallucinated content creates significant barriers in applications requiring factual accuracy. Current detection methods mainly depend on internal signals like uncertainty and self-consistency checks, using th…
arXiv cs.AI
TIER_1English(EN)·Renato Vukovic, Hsien-chin Lin, Carel van Niekerk, Benjamin Ruppik, Michael Heck, Shutong Feng, Nurul Lubis, Milica Gasic·
arXiv:2609.05025v1 Announce Type: cross Abstract: Hallucination-where a language model generates outputs that are factually incorrect or unsupported by the source-is a major challenge for both prompted and fine-tuned language models. Detecting hallucinations is difficult due to t…
Hallucination-where a language model generates outputs that are factually incorrect or unsupported by the source-is a major challenge for both prompted and fine-tuned language models. Detecting hallucinations is difficult due to the opaque reasoning processes of LLMs, which often…
arXiv:2609.03953v1 Announce Type: new Abstract: Understanding the frequency of factual errors in chatbot-generated text and evaluating systems that detect these errors is critical for determining chatbot safety. Yet factual-error detection is often treated as a single-pass, singl…
arXiv:2609.03580v1 Announce Type: new Abstract: The growing scale of academic peer review has motivated the use of Large Language Models (LLMs) as review assistants, yet LLMs can generate fluent but unsupported claims that undermine review reliability. Existing hallucination benc…
Understanding the frequency of factual errors in chatbot-generated text and evaluating systems that detect these errors is critical for determining chatbot safety. Yet factual-error detection is often treated as a single-pass, single-annotator labeling problem. In long-form chatb…
The growing scale of academic peer review has motivated the use of Large Language Models (LLMs) as review assistants, yet LLMs can generate fluent but unsupported claims that undermine review reliability. Existing hallucination benchmarks are not designed for peer review, where v…
arXiv:2609.02731v1 Announce Type: cross Abstract: Large vision-language models have achieved remarkable success in vision-language tasks. However, they remain prone to Visual Hallucinations (VHs), undermining their reliability in real-world applications. Existing solutions typica…
arXiv:2609.02679v1 Announce Type: cross Abstract: When LLMs support public-facing or high-stakes workflows, missed fabrications can harm users and institutions, while false alarms consume limited human-review capacity. When no trusted context or reference document is available, w…
arXiv cs.AI
TIER_1English(EN)·Wooje Park, Insu Lee, Soohyun Kim, Jaeyun Jang, Minyoung Noh, Kyuhong Shim, Byonghyo Shim·
arXiv:2603.23934v2 Announce Type: replace-cross Abstract: Large vision-language models (LVLMs) are increasingly being applied to multi-view image inputs captured from diverse viewpoints. Despite this growing use, current LVLMs often generate incorrect responses due to visual inte…
arXiv cs.CL
TIER_1English(EN)·Elisei Rykov, Timur Ionov, Nikolay Ivanov, Maksim Savkin, Maksim Makarenko, Alexander Panchenko, Vasily Konovalov, Julia Belikova·
arXiv:2609.00581v1 Announce Type: new Abstract: Ensuring factuality remains a critical challenge for deploying LLMs in high-stakes settings. Existing hallucination detectors usually operate at a single level: claim-level methods provide interpretable factual units, while span-lev…
arXiv:2510.12851v2 Announce Type: replace-cross Abstract: Large Audio-Language Models (LALMs) excel in Audio QA but often suffer from hallucinations ungrounded in the audio. To our knowledge, we are the first to propose applying vector steering to the audio domain to mitigate thi…
Despite recent advances in large vision-language models (LVLMs), object hallucination remains a major barrier to their reliable deployment. Existing detection methods often characterize visual grounding using attention from individual layers, leaving its evolution across layers u…
arXiv:2608.29924v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) are prone to hallucinations: they fluently describe objects, attributes, and scenes that are not in the image. We connect part of this failure to a measurable property of their representations,…
arXiv:2608.29092v1 Announce Type: new Abstract: Large vision-language models (LVLMs) frequently generate content unsupported by visual inputs. Preliminary experiments show that visual evidence is primarily incorporated into answer-side representations in early-to-middle decoder l…
arXiv cs.LG
TIER_1English(EN)·Afsaneh Hasanebrahimi, Hanxun Huang, Christopher Leckie, Sarah Erfani·
arXiv:2608.30480v1 Announce Type: cross Abstract: Object hallucination remains a persistent reliability issue in large vision-language models, where generated object mentions may sound plausible but lack visual grounding. Recent training-free detectors use internal signals such a…
arXiv:2608.29974v1 Announce Type: cross Abstract: Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at identifying hallucinated text spans but suffer fr…
arXiv:2602.11166v2 Announce Type: replace-cross Abstract: Parameter-efficient fine-tuning (PEFT) methods are widely used to adapt large language models (LLMs) to downstream tasks and are often assumed to improve factual correctness. However, how the parameter-efficient fine-tunin…
arXiv cs.AI
TIER_1English(EN)·Joonki Min, Chaeyun Kim, Hyungwook Choi, Yejin Kim, Kihyun Kim, Yohan Jo, Joonseok Lee·
arXiv:2608.30653v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) are increasingly deployed in multi-image scenarios requiring complex reasoning across visual contexts. However, current MLLMs remain fundamentally limited by object hallucination-generating…
arXiv:2608.29307v1 Announce Type: cross Abstract: Language models increasingly answer questions by consulting retrieved documents rather than memory alone, a design now common in search assistants and enterprise knowledge tools. Grounding a model in retrieved text reduces unsuppo…
Enoki is an open information extraction framework that unifies claim-level verification and span-level hallucination localization through shared relational facts, reducing resource use while improving detection accuracy.
Multimodal Large Language Models (MLLMs) are increasingly deployed in multi-image scenarios requiring complex reasoning across visual contexts. However, current MLLMs remain fundamentally limited by object hallucination-generating plausible yet factually inconsistent descriptions…
Detecting hallucinations in Large Vision-Language Models (LVLMs) requires both accurate span localization and well-calibrated confidence scores. Fine-tuned generative VLMs excel at identifying hallucinated text spans but suffer from overconfidence and high inference latency. Disc…
A hybrid system combining a multimodal sequence tagger and a generative vision-language model improves hallucination span detection and calibration through union-calibrated fusion.
arXiv:2609.17800v1 Announce Type: new Abstract: Vision-language models (VLMs) remain largely unreliable on panoramic dental radiographs and can rely on learned anatomical priors rather than evidence in the image. This is particularly problematic for tooth localization and spatial…
arXiv:2609.16601v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have made strong progress on visual question answering and image captioning, yet they still produce fluent claims about objects, attributes, or relations that are not grounded in the image. M…
arXiv:2609.16646v1 Announce Type: new Abstract: When strong multimodal models are widely available, progress requires new scientific methodologies beyond benchmark scores---using models as instruments for understanding behavior. We address this by asking: can we use large vision-…
arXiv cs.CV
TIER_1English(EN)·Xinyu Chen, Adnan Mahmood, Mark Dras·
arXiv:2609.09895v1 Announce Type: new Abstract: Video-language models and video agents can produce hallucinations that conflict with spatiotemporal evidence. Existing benchmarks mainly evaluate model hallucinations, and heterogeneous mechanisms make detector reliability difficult…
arXiv:2609.02028v1 Announce Type: new Abstract: Despite recent advances in large vision-language models (LVLMs), object hallucination remains a major barrier to their reliable deployment. Existing detection methods often characterize visual grounding using attention from individu…
arXiv cs.CV
TIER_1English(EN)·Mehrdad Fazli, Sina Mansouri, Mohit Marvania, Ziwei Zhu·
arXiv:2609.00231v1 Announce Type: new Abstract: Existing research on object hallucination in multimodal large language models (MLLMs) predominantly attributes the problem to language priors such as over-reliance on textual co-occurrence statistics. We challenge this view by prese…
arXiv cs.CV
TIER_1English(EN)·Kairong Yu, Zixin Zhu, Le Yu, Hongwei Wang·
arXiv:2608.28058v1 Announce Type: new Abstract: Large Vision-Language Models (LVLMs) remain prone to hallucinations, producing responses that are irrelevant or inconsistent with the multimodal input. Existing mitigation methods mainly rely on external supervision, output calibrat…
dev.to — Claude Code tag
TIER_1English(EN)·golproductions·
<p>AI hallucination has two sides: what the model knows, and what the model can check.</p> <p>Knowledge is hard. The best frontier model in 2026 scores 43 out of 100 on open-domain factual knowledge. Reasoning models made it worse. o3 hallucinates at 33% compared to o1's 16%. Nob…
dev.to — MCP tag
TIER_1English(EN)·Programming Central·
<p>Artificial intelligence has a trust problem. For years, we have danced between two contrasting paradigms: the statistical, probabilistic pattern-matching of deep neural networks and the rigorous, deterministic rule-processing of symbolic logic systems. Modern Large Language Mo…
