PARALLAX: Separating Genuine Hallucination Detection from Benchmark Construction Artifacts

• URL: http://arxiv.org/abs/2605.17028v1
• Date: Sat, 16 May 2026 14:57:15 GMT
• Title: PARALLAX: Separating Genuine Hallucination Detection from Benchmark Construction Artifacts
• Abstract summary: Large language models (LLMs) hallucinate with confidence.<n> detecting it from internal model states offers a path to safer deployment.<n>Recent methods achieving high detection performance on widely used benchmarks.<n>We show that much of this apparent progress does not survive scrutiny.
• Score: 12.088141798203367
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models (LLMs) hallucinate with confidence: their outputs can be fluent, authoritative, and simply wrong. In medical, legal, and scientific applications this failure causes direct harm, and detecting it from internal model states offers a path to safer deployment. A growing body of work reports that this problem is increasingly tractable, with recent methods achieving high detection performance on widely used benchmarks. We show, however, that much of this apparent progress does not survive scrutiny. Four of the six corpora embed the ground-truth answer directly in the input prompt. A naïve text-similarity baseline we call \textsc{TxTemb} exploits this to achieve near-perfect detection scores without any access to model internals. To measure what genuine detection capability remains once these artifacts are controlled, we conduct a large-scale evaluation spanning twenty-two detection methods, twelve open-source models spanning six architectural families, and six corpora. We further introduce \textbf{DRIFT}, a supervised probe over inter-layer hidden-state transitions, as a point of comparison for live-generation detection. Our findings suggest that the field's reported progress on hallucination detection is substantially explained by benchmark construction artifacts in widely used corpora, and that the majority of established baselines perform near chance under controlled conditions; the consistent exceptions are SAPLMA and DRIFT, both supervised probes on upper-layer hidden states.

Related papers

• Actionable Hallucination Detection: Translating Latent Uncertainty into Agentic Critique [1.1172382217477128]
  Large Language Models (LLMs) deployed as AI agents frequently exhibit user specification-grounding failures.<n>Existing detection methods fail to provide actionable, real-time correction as they either do not localize the hallucinations, or incur prohibitive inference latency.<n>We introduce the Latent Critic, a lightweight low-rank adapter (LoRA) that operates concurrently with a frozen base LLM's generation.
  arXiv  Detail & Related papers  (2026-08-11T03:26:16Z)
• From Signals to Transfer: A Factorised Study of Probe-Based Uncertainty Estimation in Large Language Models [51.470637128942336]
  We propose a factorised study of probe-based uncertainty estimation (UE) under matched conditions.<n>Our results show that raw hidden states and attention features are difficult to outperform in-domain.<n>We train benchmark-based pretrained probes that transfer reasonably well to open-ended factual generation.
  arXiv  Detail & Related papers  (2026-06-26T03:23:15Z)
• "Did you lie?" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms [2.3883252067956118]
  We show that existing trained model organisms often fail testbeds where models verifiably believe the opposite of what they say.<n>We address this with 13 reasoning model organisms whose hidden beliefs are verified in chain-of-thought.<n>On prompted lying, across 31 open-weight models spanning 2B to 1T parameters, all four detectors show positive scaling with model capability.
  arXiv  Detail & Related papers  (2026-06-10T19:21:12Z)
• Beyond Final Answers: Auditing Trajectory-Level Hallucinations in Multi-Agent Industrial Workflows [1.6048497142263283]
  Trajel is a dataset and evaluation framework for auditing trajectory-level hallucinations in multi-agent agents.<n>We benchmark supervised detection models at the subtask, trajectory, and long-context levels.<n>Trajectory-aware detection significantly outperforms standard post-hoc verification.
  arXiv  Detail & Related papers  (2026-05-22T21:05:10Z)
• Sanity Checks for Long-Form Hallucination Detection [6.501493411588847]
  Hallucination detection methods for large language models operate on chain-of-thought reasoning traces, yet it remains unclear whether they evaluate the reasoning itself or merely exploit surface correlates of the final answer.<n>We introduce a controlled-invariance methodology that exposes this distinction through two oracle tests: textscForce, which replaces each response's final answer with the ground truth while preserving the reasoning trace, and textscRemove, which strips answer-announcement steps while leaving the trajectory intact.<n>This reveals if their predictive power derives from answer-level artifacts rather than from the structure
  arXiv  Detail & Related papers  (2026-05-08T18:00:58Z)
• ExposeAnyone: Personalized Audio-to-Expression Diffusion Models Are Robust Zero-Shot Face Forgery Detectors [58.45131932883374]
  We propose a fully self-supervised approach to detect deepfakes in videos.<n>Our model computes the identity distances between suspected videos and personalized subjects via diffusion reconstruction errors.<n>Our method is highly robust to corruptions such as blur and compression, highlighting the applicability in real-world face forgery detection.
  arXiv  Detail & Related papers  (2026-01-05T18:59:54Z)
• Sparse deepfake detection promotes better disentanglement [4.901409400999413]
  We show that sparse deepfake detection can improve detection performance, with an EER of 23.36% on ASVSpoof5 test set, with 95% of sparsity.<n>We then show that these representations provide better disentanglement, using completeness and modularity metrics based on mutual information.
  arXiv  Detail & Related papers  (2025-10-07T09:03:39Z)
• Why Speech Deepfake Detectors Won't Generalize: The Limits of Detection in an Open World [11.238970239267248]
  Speech deepfake detectors are often evaluated on clean, benchmark-style conditions.<n>But deployment occurs in an open world of shifting devices, sampling rates, codecs, environments, and attack families.<n>This creates a coverage debt" for AI-based detectors, producing data blind spots that grow faster than data can be collected.
  arXiv  Detail & Related papers  (2025-09-23T20:27:04Z)
• Beyond Artificial Misalignment: Detecting and Grounding Semantic-Coordinated Multimodal Manipulations [56.816929931908824]
  We pioneer the detection of semantically-coordinated manipulations in multimodal data.<n>We propose a Retrieval-Augmented Manipulation Detection and Grounding (RamDG) framework.<n>Our framework significantly outperforms existing methods, achieving 2.06% higher detection accuracy on SAMM compared to state-of-the-art approaches.
  arXiv  Detail & Related papers  (2025-09-16T04:18:48Z)
• Detecting High-Stakes Interactions with Activation Probes [10.421494494629421]
  This paper examines activation probes for detecting "high-stakes" interactions, where the text indicates that the interaction might lead to significant harm.<n>We evaluate several probe architectures trained on synthetic data, and find them to exhibit robust generalization to diverse, out-of-distribution, real-world data.<n>Our experiments also highlight the potential of building resource-aware hierarchical monitoring systems.
  arXiv  Detail & Related papers  (2025-06-12T15:20:33Z)
• Unifying Unsupervised Graph-Level Anomaly Detection and Out-of-Distribution Detection: A Benchmark [73.58840254552656]
  Unsupervised graph-level anomaly detection (GLAD) and unsupervised graph-level out-of-distribution (OOD) detection have received significant attention in recent years.<n>We present a underlinetextbfUnified underlinetextbfBenchmark for unsupervised underlinetextbfGraph-level underlinetextbfOOD and anomaunderlinetextbfLy underlinetextbfDetection (ourmethod)<n>Our benchmark encompasses 35 datasets
  arXiv  Detail & Related papers  (2024-06-21T04:07:43Z)
• DF40: Toward Next-Generation Deepfake Detection [62.073997142001424]
  existing works identify top-notch detection algorithms and models by adhering to the common practice: training detectors on one specific dataset and testing them on other prevalent deepfake datasets. But can these stand-out "winners" be truly applied to tackle the myriad of realistic and diverse deepfakes lurking in the real world? We construct a highly diverse deepfake detection dataset called DF40, which comprises 40 distinct deepfake techniques.
  arXiv  Detail & Related papers  (2024-06-19T12:35:02Z)
• A Comprehensive Library for Benchmarking Multi-class Visual Anomaly Detection [89.92916473403108]
  This paper proposes a comprehensive visual anomaly detection benchmark, ADer, which is a modular framework for new methods.<n>The benchmark includes multiple datasets from industrial and medical domains, implementing fifteen state-of-the-art methods and nine comprehensive metrics.<n>We objectively reveal the strengths and weaknesses of different methods and provide insights into the challenges and future directions of multi-class visual anomaly detection.
  arXiv  Detail & Related papers  (2024-06-05T13:40:07Z)
• A New Benchmark and Reverse Validation Method for Passage-level Hallucination Detection [63.56136319976554]
  Large Language Models (LLMs) generate hallucinations, which can cause significant damage when deployed for mission-critical tasks. We propose a self-check approach based on reverse validation to detect factual errors automatically in a zero-resource fashion. We empirically evaluate our method and existing zero-resource detection methods on two datasets.
  arXiv  Detail & Related papers  (2023-10-10T10:14:59Z)
• SeeABLE: Soft Discrepancies and Bounded Contrastive Learning for Exposing Deepfakes [7.553507857251396]
  We propose a novel deepfake detector, called SeeABLE, that formalizes the detection problem as a (one-class) out-of-distribution detection task. SeeABLE pushes perturbed faces towards predefined prototypes using a novel regression-based bounded contrastive loss. We show that our model convincingly outperforms competing state-of-the-art detectors, while exhibiting highly encouraging generalization capabilities.
  arXiv  Detail & Related papers  (2022-11-21T09:38:30Z)

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.