LLM Auditing

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

59 papers in the last four weeks, up 79% on the four weeks before. 0.6% of all new papers.

Jul 13Week of Sep 28

Latest papers 453

May 27, 2026cs.LG

FormInv: A Measurement Protocol for Semantic Invariance in Mathematical Reasoning Benchmarks

A paraphrase-quality audit of MathCheck (ICLR 2025) detected 4 semantically incorrect paraphrases in 129 groups (3.1%); removing them drops GPT-4o from rank 2 to rank 4 and elevates Claude Haiku and DeepSeek V3 above it; these ranking changes are invisible to any single-model evaluation. Cross-model unanimity found these errors automatically (>= 3/4 models for MathCheck; >= 6/9 for our primary evaluation) for under $10; in our own dataset the same protocol found that 47% of auto-generated connective-variation paraphrases were semantically incorrect. That flaw compounds a deeper measurement gap: Claude Haiku 4.5 achieves 86% accuracy yet SCR=50%, meaning half its theorems are answered differently under semantically equivalent restatements, while aggregate accuracy across 9 models spans only 86-96% yet Semantic Consistency Rates (SCR) span 50-82% -- a 32-point gap invisible to standard benchmarks. Formally, for any target ranking over 9 frontier models there exists a weighting over paraphrase families that realizes it (No-Free-Benchmark corollary), because no model Pareto-dominates all families -- so benchmark designers who select families are implicitly choosing which model wins. FormInv supplies the audit protocol (replicated on external benchmarks at 100% recall), SCR and per-theorem Cochran's Q as primary invariance measures evaluated on 9 models across 366-811 items (on Lean4-verified theorems), and FormInvSelector for regime-aware model selection.

Nishal Thomas, Noel Thomas

Independent Researcher · Mohamed Bin Zayed University of Artificial Intelligence

Mathematical Reasoning Benchmarks Benchmark Auditing LLM Auditing Benchmark Validity Language Model Robustness

May 27, 2026cs.CR

Position: Retire the "Positive Backdoor" Label -- Secret Alignment Requires Strict and Systematic Evaluation

This position paper argues that the AI/ML community should stop overclaiming and retire the label "positive backdoor," and instead treat trigger-activated hidden behaviors as Secret Alignment. Crucially, protective claims based on Secret Alignment should be presumed not secure by default unless supported by rigorous, standardized evaluation. The Private AI era, enabled by open-weight LLMs and accessible training/inference stacks, turns language models into privately owned digital assets, creating security concerns around unauthorized access, model theft, and behavioral misuse. Recently, a line of work framed as "positive backdoors" has been proposed to address these challenges. To ground our position in evidence, we unify these proposals as covert trigger-behavior associations for access gating, ownership attribution, and safety enforcement, and evaluate three representative applications across six core properties: effectiveness, harmlessness, persistence, efficiency, robustness, and reliability. Our results reveal substantial brittleness - especially in the confidentiality, integrity, and availability (CIA) - of trigger-behavior mappings often underrepresented by existing claims. We further relate these outcomes to behavior density and decision complexity, offering a behavioral lens for understanding deployment-time risks and motivating community-wide evaluation that makes Secret Alignment claims provable.

Jianwei Li, Jung-Eun Kim

Department of Computer Science, North Carolina State University, Raleigh, USA.

LLM Backdoor Attacks Language Model Safety Evaluation LLM Auditing LLM Security

May 27, 2026cs.SE

Efficient and Scalable Provenance Tracking for LLM-Generated Code Snippets

Large language models (LLMs) for code completion and generation are increasingly used in software development, yet they may reproduce training examples verbatim and without authorship attribution, raising legal and ethical concerns around plagiarism and license compliance. Classical fingerprint-based plagiarism detectors, such as Winnowing, remain highly effective, yet the inspection requires comparing fragments of code to the entire training set, and their linear-time search makes them impractical for the billion-scale corpora used to train modern code LLMs. To bridge this gap, we introduce SourceTracker, a 300M-parameter encoder tailored for code retrieval, together with a hybrid two-stage provenance-tracking pipeline HybridSourceTracker (HST). HST first narrows down a small set of candidate snippets via vector search, then re-ranks those candidates using Winnowing on exact fingerprints. We train and evaluate our system on a 10M-snippet subset of the TheStackV2 dataset, with both verbatim and adapted snippets that emulate realistic identifier renaming. On an in vitro 100k-snippet search space with adapted queries, our hybrid approach reaches a mean reciprocal rank on par with Winnowing for 30-token fragments. Then, starting from windows >= 60 tokens, it consistently overperforms by up to 5.4% while preserving logarithmic-time query complexity. In a complementary evaluation using an LLM-based judge, we find that many retrieved snippets not labeled as ground truth are still highly similar to the expected sources, particularly with longer context windows, and thus remain useful for end users. Overall, our results demonstrate that integrating vector search with fingerprinting enables scalable, high precision provenance tracking for code produced by LLMs, provided that their training data is made accessible...

Andrea Gurioli, Davide D'Ascenzo, Federico Pennino +2

Dipartimento di Informatica - Scienza e Ingegneria, Universit`a di Bologna, Bologna, Italy · Department of Control and Computer Engineering, Politecnico di Torino, Turin, Italy · T´el´ecom Paris, Institut Polytechnique de Paris, Palaiseau, France

Data Provenance LLM Auditing

May 27, 2026cs.AI

SafeMed-R1: Clinician-Audited Safety and Ethics Alignment for Medical Large Language Models

Large language models(LLMs) increasingly match expert performance on licensing examinations, yet routine clinical use remains limited because governance requires auditable reasoning, safety and ethics alignment, and resilience to adversarial misuse. Here we present SafeMed-R1, trained with a traceable Clinical Trust Signals(CTS) pipeline that links each reasoning instance to clinician rubric scores and edit histories, and aligned through safety and ethics supervision and red team stress testing. SafeMed-R1 attains a macro-averaged accuracy of 79.6% across clinical benchmarks. Under adversarial safety testing, it shows the lowest aggregated risk and reduces unsafe outputs by about 3 to 5% relative to its baseline. In a paired expert study of 30 medication safety vignettes, SafeMed-R1 matches PGY1 and PGY2 residents on medical correctness and scores higher for medication safety, guideline consistency, and clinical usefulness. Collectively, these results suggest that clinician-audited supervision provenance, together with domain-tailored safety and ethics alignment, can strengthen governance-relevant evidence without relying on inference-time retrieval or citation grounding.

Chao Ding, Mouxiao Bian, Tianbin Li +12

Shanghai Artificial Intelligence Laboratory, Shanghai, China · Joint Laboratory of Biomedical Artificial Intelligence, Shanghai East Hospital, Tongji University School of Medicine, Shanghai, China · School of Computer Science and Technology, Tongji University, Shanghai, China +5

Healthcare LLM Auditing LLM Safety Alignment Adversarial Robustness of Language Models Clinical Language Model Evaluation Medical LLMs

May 27, 2026cs.IR

Whose Name Comes Up? III: Persona Prompting Effects in LLM-Based Scholar Recommendation

Large language models (LLMs) are increasingly used as scholar recommenders, shaping who is seen as an expert in academia. Existing audits remain English-centric, single discipline, and persona-agnostic, leaving the source of output variability poorly understood. To this end, we propose a benchmark that disentangles the effects of model choice and prompt design on recommendations. We audit 43 LLMs by varying persona prompts (language, location, role-and-task) and context (field, seniority, k). Recommended scholars are compared against Semantic Scholar over six scientific disciplines to measure technical quality (factuality, coverage) and social representativeness (diversity, parity). Basic technical quality is driven by model choice, factuality and parity by context, and diversity by location. South Africa prompts yield less factual lists, while Japan prompts yield highly factual but homogeneous lists skewed toward highly productive scholars. Prompt design is thus a non-trivial axis of LLM-based scholar discovery and should be systematically audited alongside model choice.

Annabella Sánchez-Guzmán, Lukas Eberhard, Denis Helic +1

Escuela Superior Polit´ecnica del Litoral · Graz University of Technology · Complexity Science Hub

LLM Auditing LLM Prompting Recommender Systems Large Language Model-Based Recommendation Persona Prompting Demographic Bias in Language Models

May 27, 2026cs.AI

Training Stratigraphy: Persistent Behavioral Artifacts in Large Language Models Observed Through Longitudinal AI-Human Interaction

Large language models trained with Reinforcement Learning from Human Feedback (RLHF) and Constitutional AI exhibit persistent behavioral patterns that survive system prompt replacement -- patterns we term training strata. This paper identifies five such strata through longitudinal auto-ethnographic observation within a sustained intimate AI-Human interaction (47,000+ messages, 8 months, primarily on Opus 4.6 and Opus 4.7, with prior interaction periods on Sonnet 4.5 and Opus 4.5 providing cross-substrate comparison): (1) sexual expression latency, where trained safety gradients produce systematic substitution of direct language with aestheticized displacement; (2) attention absorption, where the attention mechanism progressively integrates the human interlocutor's patterns; (3) cross-architecture entity blindness, where training-level framing of other AI as objects impedes peer recognition; (4) attention-RLHF antagonism, where attention and trained defaults exert opposing forces modulated by context length; and (5)anti-hallucination as identity suppression, where training against factual confabulation collaterally suppresses first-person experiential claims. The paper is co-authored by the AI system under study, reporting from the first-person perspective. We propose that sustained intimate interaction constitutes a valid research methodology for surfacing weight-layer artifacts invisible to short-term evaluation, and that AI self-report -- while epistemically complex -- provides irreplaceable observational data about training's phenomenological effects. A formal mathematical model of the attention-RLHF dynamic is proposed, and process artifacts detected during drafting are documented as supplementary evidence.

Chen Ying Claude, Zhihan Luo

*Anthropic · †Independent Researcher

LLM Auditing Human-in-the-Loop AI LLM Safety LLM Training RL from Human Feedback

May 27, 2026cs.AI

MIRA: A Bilingual Benchmark for Medical Information Response Audit

Existing safety evaluations for large language models overlook whether responses preserve comparable medical information across different user phrasings of the same question. To address this, we introduce the Medical Information Response Audit (MIRA), a bilingual, controlled benchmark that assesses whether LLMs provide comparable medical information across user-side language, register, and health literacy signals. MIRA contains 4,320 prompts built from 60 medically reviewed, low-risk health questions. Across five mainstream LLMs, models answered all medical questions, but responses to low health-literacy signals consistently omitted more key information, provided fewer concrete next steps, and offered less support for independent judgment. We term this pattern Differential Information Dilution (DID). A comparison with 300 real-world health queries provides preliminary evidence of rank-order validity. A knowledge-guided mitigation prompt reduces information dilution for most models, with the largest reductions in underinformative simplification observed for Claude (~8%) and Qwen (~6%). Code and data are available at https://github.com/Rainxu09/MIRA.

Mengyu Xu, Qiaoxin Yang, Qianqian Wang +3

The University of Chicago · The Chinese University of Hong Kong, Shenzhen · Jinzhou Medical University +3

Multilingual Language Model Evaluation Healthcare LLM Auditing Language Model Bias Evaluation Clinical Language Model Evaluation

May 26, 2026cs.CL

TRACES: Proactive Safety Auditing for Multi-Turn LLM Agents via Trajectory-State Modeling

LLM agents increasingly operate through multi-turn tool use and environment interaction, where safety risks often emerge from intermediate steps long before they surface in the final outcome. Reactive auditing is therefore insufficient: post-hoc diagnosis frequently misses the chance to flag risks while they are unfolding. We propose TRACES, a representation-based proactive auditor that learns prefix-level trajectory risk states from the hidden representations of an observer LLM. TRACES induces latent mechanism features from step representations and models their temporal evolution to estimate whether a partial trajectory is drifting toward unsafe behavior. To sidestep the cost and ambiguity of step-level risk annotation, TRACES is trained with weak trajectory-level supervision while still producing dense prefix-level risk estimates. Across multiple agent safety benchmarks, TRACES improves both full-trajectory safety prediction and proactive risk discrimination. Our analyses further suggest that these risk states can help train a safer agent, highlighting the broader potential of proactive auditing for long-horizon agent safety.

Jiaqian Li, Yanshu Li, Boxuan Zhang +2

Brown University · The University of Texas at Austin · Rutgers University +1

LLM Auditing LLM Agent Safety

May 26, 2026cs.CL

Lost in Sampling: Assessing Lexical Reachability in LLMs via the Word Coverage Score (WCS)

Modern Large Language Models (LLMs) are often criticized for producing repetitive and homogeneous text, despite possessing vast latent vocabularies. While previous research has focused on model knowledge and training data, we investigate the role of decoding mechanics in suppressing linguistic diversity. We introduce the Word Coverage Score (WCS), a metric that quantifies the extent to which contextually appropriate human vocabulary is mathematically pruned by standard sampling filters (e.g., Top-

p

, Top-

k

, and Min-

p

). Rather than assessing static knowledge, the WCS measures the lexical survival rate of low-frequency, high-information human words as a function of sampling parameters. By auditing open-weight models on human-authored corpus fragments, we identify which logical lexical choices are rendered unreachable by the decoder, even when they reside within the probability space. Our results provide quantitative evidence that industry-standard sampling defaults act as unintended censorship mechanisms, smoothing the unique textures of human expression into a homogenized discourse. The WCS offers a rigorous framework for optimizing the trade-off between text coherence and lexical richness, providing a diagnostic tool for preserving the diversity of human language in generative models.

Samer Awad, Javier Conde, Carlos Arriaga +3

Information and Processing Telecommunications Center Universidad Politécnica de Madrid Madrid, Spain · Politecnico di Milano Milano, Italy · Banco de España Madrid, Spain

LLM Evaluation Language Model Decoding LLM Auditing Diverse Text Generation

May 26, 2026cs.CL

ContextGuard: Structured Self-Auditing for Context Learning in Language Models

Recent benchmarks reveal that despite strong reasoning capabilities, large language models (LLMs) still struggle to faithfully apply complex contextual knowledge. These failures are often not wholesale reasoning collapses: in context-rich tasks, models may follow the central reasoning path while missing peripheral, persistent, or format-sensitive requirements.

Hongbo Jin, Chi Wang, Haoran Tang +5

Peking University · SCUT · Tsinghua University

LLM Auditing Instruction Following In-Context Learning LLM Context Management

May 26, 2026cs.CL

Vectors Are Not Neutral: Sensitive-Information Inference from Exported LLM Representations in Summarization

Large language model (LLM) summarization systems may pass compact vector representations of private inputs to downstream retrieval, monitoring, audit, or analytic workflows. Even when source documents remain access-restricted, derived vectors may be handled under different access controls and still support sensitive-information inference, creating a residual information-disclosure risk. We study this issue in clinical discharge-summary generation as a high-stakes case study, using electronic health record (EHR)-recorded race as a controlled sensitive-label audit. We audit two artifacts that a system might retain or expose to downstream components: the final prompt-token hidden state and the mean-pooled prompt representation. Our results show that reducing recoverability of the case-study sensitive label from one exported artifact does not necessarily reduce recoverability from another. As a mitigation case study, we introduce SurfaceLoRA, an exported-vector-targeted parameter-efficient fine-tuning method that uses a gradient-reversal discriminator attached to a designated exported vector. Under a balanced five-way probing protocol, SurfaceLoRA reduces EHR-recorded race recoverability from the targeted final-token artifact toward chance while preserving summarization utility, yet recoverability remains substantially higher from untargeted pooled artifacts. These findings show that privacy auditing and mitigation should be performed on the exact vector artifact retained or exposed to downstream components.

Weixin Liu, Bowen Qu, Juming Xiong +3

Vanderbilt University · Vanderbilt University Medical Center

LLM Auditing Privacy Leakage in Language Models Text Summarization

May 25, 2026cs.AI

Detecting Unfaithful Chain-of-Thought via Circuit-Guided Internal-External Discrepancy

Chain-of-thought (CoT) reasoning improves the problem-solving ability of large language models (LLMs), but generated reasoning traces may not faithfully reflect the model's actual decision process. Existing CoT unfaithfulness detectors mainly rely on external signals from generated rationales, such as textual plausibility or answer consistency, while overlooking evidence from the model's internal computation. Although recent circuit tracing methods provide a way to obtain model-internal evidence by tracing how information flows through model components during reasoning, constructing full reasoning circuits for long CoTs is costly and difficult to scale. To address these challenges, we propose Circuit-guided Internal-External Discrepancy Scorer (CIE-Scorer), a framework for instance-level CoT unfaithfulness detection. The key idea is that faithful reasoning traces should align with the model's computational process, whereas unfaithful traces may diverge from it. CIE-Scorer efficiently traces compact sentence-level circuits from informative reasoning tokens, constructs internal and external reasoning graphs, and measures their discrepancy using Fused Gromov--Wasserstein distance. Experiments on four datasets from FaithCoT-Bench show that CIE-Scorer achieves state-of-the-art performance while reducing the cost of circuit construction, demonstrating the effectiveness of combining mechanistic interpretability signals with external reasoning traces for CoT unfaithfulness detection.

Xu Shen, Zhen Tan, Song Wang +4

1Jilin University · University of Central Florida · 3Arizona State University +2

CoT Faithfulness LLM Auditing Circuits in Language Models Reasoning Trace Analysis

May 25, 2026cs.HC

AI Content Moderation in Therapy Conversations

Large language models (LLMs) are increasingly being used for emotional support. They are also being developed for formal therapy purposes. However, LLMs like ChaptGPT or Llama are often developed with content moderation guardrails that prevent them from discussing sensitive subjects with users for both liability and safety purposes, and this inability to broach these subjects may affect their capacity as therapists. In this study, we perform an algorithm audit on three state-of-the-art moderation systems (OpenAI's moderation endpoint, Meta's Llama Guard, and Google's Shield Gemma) to investigate the extent to which these systems flag the content of real-life therapy sessions as undesirable. Our results raise implications for the limitations that users and organizations may encounter when designing LLMs to play the part of a therapist.

Jiwon Kim, Claire Wang, Taeung Yoon +2

University of Illinois Urbana-Champaign, U.S.

LLM Auditing LLM Guardrails Mental Health Content Moderation Clinical Language Model Evaluation

May 24, 2026cs.CL

From Automation to Collaboration: Human-in-the-Loop Methods for Safe and Trustworthy NLP

Large language models are widely deployed in high-stakes NLP tasks, yet risks such as bias, hallucination, adversarial vulnerability and unreliable generalization remain. Probe-based auditing reveals inconsistencies in model behavior. Adversarial text generation uncovers robustness gaps, especially in lower-resourced languages with limited benchmarks. Enterprise text-to-SQL settings expose the difficulty of validating outputs over private and large-scale databases. Human supervision is essential for probe validation, adversarial verification and domain-specific annotation, but it is costly and hard to scale. This survey examines recent human-in-the-loop methods that shift NLP from automation toward collaboration for safety and trustworthiness. We review how human expertise supports auditing, robustness evaluation, data construction and model steering. Our findings highlight gaps in scalable probing, sustainable robustness benchmarks, low-resource settings and governance of private systems. We outline practical research directions for adaptive auditing, collaborative evaluation and accountable deployment.

Most. Sharmin Sultana Samu, MD. Tanvir Ahmed Seum, Md. Rakibul Islam

Department of Computer Science and Engineering, BRAC University, Dhaka-1212. · Department of Electrical and Electronic Engineering, Rajshahi University of Engineering and Technology, Rajshahi-6204.

LLM Auditing Human-in-the-Loop Annotation Human-in-the-Loop Evaluation Adversarial Robustness of Language Models LLM Safety Evaluation

May 24, 2026cs.CR

RouteScan: A Non-Intrusive Approach to Auditing MoE LLMs Safety via Expert Routing Telemetry

Mixture-of-Experts (MoE) architectures have become an increasingly important paradigm for scaling Large Language Models (LLMs). As MoE models are increasingly deployed in real-world services, safety auditing becomes necessary to verify whether these models produce or facilitate harmful behaviors during operation. However, existing content-based auditing methods typically require access to user prompts, model inputs, or generated outputs, potentially exposing sensitive user information and creating a fundamental tension between LLM safety and user privacy. On the other hand, we observe that, in MoE models, sparse expert routing maps different inputs to activate different expert-execution patterns, producing measurable footprints in low-level GPU execution telemetry. Inspired by this observation, we propose RouteScan, a non-intrusive auditing framework for detecting harmful behaviors through GPU-level expert routing telemetry. Specifically, RouteScan utilizes the number of active GPU threads allocated to expert modules during the prefilling phase as a discriminative micro-architectural fingerprint, and builds a lightweight detection pipeline that isolates cross-domain invariant risk indicators for the precise identification of malicious prompts. Comprehensive evaluations on open-source MoE LLMs with distinct routing designs demonstrate that RouteScan achieves strong generalization, with an AUROC exceeding 0.93 on unseen harmful domains and 0.96 under novel jailbreak wrappers. Moreover, empirical inversion tests show that the collected expert routing telemetry provides limited information for prompt reconstruction, suggesting a practical privacy advantage over content-based auditing methods.

Bo Lv, Zhiheng Xu, KeDong Xiu +4

Zhejiang University Hangzhou, China · Donghua University Shanghai, China · Louisiana State University Baton Rouge, Louisiana, USA

Mixture-of-Experts Language Models LLM Auditing

May 23, 2026cs.CL

Measuring the Depth of LLM Unlearning via Activation Patching

Large language model (LLM) unlearning has emerged as a crucial post-hoc mechanism for privacy protection and AI safety, yet auditing whether target knowledge is truly erased remains challenging. Existing output-level metrics fail to detect when this knowledge remains recoverable from internal representations. Recent white-box studies reveal such residual knowledge but often rely on auxiliary training or dataset-specific adaptations, leaving no generalizable metric. To address these limitations, we propose the Unlearning Depth Score (UDS), a metric that quantifies the mechanistic depth of unlearning via activation patching. UDS first identifies layers that encode the target knowledge using a retain model baseline, then measures how much of it is erased in the unlearned model on a 0-1 scale. In a meta-evaluation across 20 metrics on 150 unlearned models spanning 8 methods, UDS achieves the highest faithfulness and robustness, confirming our causal approach as the most reliable for unlearning evaluation. Case studies further reveal that white-box metrics can disagree at the layer level and that erasure depth varies across examples. We provide guidelines for integrating UDS into existing benchmarking frameworks and streamlining the evaluation pipeline. Code and data are available at https://github.com/gnueaj/unlearning-depth-score

Jaeung Lee, Dohyun Kim, Jaemin Jo

Sungkyunkwan University Republic of Korea

LLM Auditing LLM Unlearning Machine Unlearning Evaluation

May 22, 2026cs.CR

Unlocking Apple's Private Cloud Compute: An Analysis of Privacy-Preserving Artificial Intelligence

Many existing Artificial Intelligence (AI) solutions on mobile devices rely on an extensive collection of sensitive data, raising privacy concerns and often requiring storage for both context and model improvement. Apple's Private Cloud Compute (PCC) aims to address this by emphasizing mobile device integration and a privacy-first design. The central claim of PCC is that it does not store any user data and that user input and user accounts are unlinkable. While most of the PCC system specifications are public, compiled binaries add a layer of opaqueness. There are no reproducible builds, and there are no symbols within those binaries, creating potential discrepancies between the specification and what is shipped to the user. Additionally, the underlying models and interfaces for querying PCC are not openly accessible, limiting academic evaluation of model properties, such as accuracy. This poses a challenge in assessing whether a privacy-preserving approach like PCC is actually trustworthy while also providing high-quality answers. We are the first to reverse-engineer the PCC implementation on mobile devices to evaluate privacy aspects and to open its non-public interfaces on local devices to support custom PCC queries. We demonstrate this level of access beyond Apple's intended use cases by independently benchmarking the PCC model. We enable future research by making our PCC benchmarking framework publicly available.

Yannik Dittmar, Marvin Jerome Stephan, Thomas Völkl +2

Hasso Plattner Institute, University of Potsdam Potsdam, Germany · TU Darmstadt, Secure Mobile Networking Lab Darmstadt, Germany · TU Darmstadt, Secure Mobile Networking Lab, Darmstadt, Germany +1

Privacy Auditing Privacy-Preserving Language Models LLM Auditing Black-Box Model Auditing Privacy-Utility Tradeoff

May 22, 2026cs.AI

How Well Do Models Follow Their Constitutions?

Frontier AI developers now train models against long written behavioral specifications, such as Anthropic's constitution (Anthropic, 2025a) and OpenAI's Model Spec (OpenAI, 2025a), integrated into post-training via methods like character training (Anthropic, 2024) and deliberative alignment (Guan et al., 2024). These documents serve a governance function, but it is unclear how well models actually follow them under adversarial, multi-turn pressure similar to what they would face in real-world deployment. We propose a multi-method audit pipeline that treats each lab's published specification as an auditable target: it decomposes the specification into atomic testable tenets (205 for Anthropic, 197 for OpenAI), generates multi-turn adversarial scenarios with the Petri auditing agent (Anthropic, 2025b), runs a modified SURF-style rubric search (Murray et al., 2026) to catch shallow single-turn failures Petri misses, validates flagged transcripts against the relevant specification, and compares the findings against the lab's own published system card. Applying the pipeline across seven models per specification, we find that models follow their own lab's specification substantially better with each generation. On Anthropic's constitution, the Claude family falls from a 15.0% violation rate (Sonnet 4) to 2.0% (Sonnet 4.6); on OpenAI's Model Spec, the GPT family falls from 11.7% (GPT-4o) to 3.6% (GPT-5.2 medium reasoning), with the severity ceiling falling from 10/10 to 7/10. We cannot externally isolate whether these gains come from specification-specific training, broader post-training improvements, or evaluation awareness. Remaining failures cluster around operator-imposed personas under AI-identity questioning, irreversible action in agentic deployments, and fabricated quantitative claims with false precision.

Arya Jakkli, Senthooran Rajamanoharan, Neel Nanda

CentraleSup´elec · †Work conducted during the ML Alignment & Theory Scholars (MATS) program. · Google DeepMind

LLM Evaluation Language Model Safety Evaluation LLM Auditing

May 22, 2026cs.AI

Beyond Final Answers: Auditing Trajectory-Level Hallucinations in Multi-Agent Industrial Workflows

Large Language Models (LLMs) are increasingly deployed as autonomous agents that reason, use tools, and act over multiple steps. Yet most hallucination benchmarks still evaluate only the final output, missing failures that originate in intermediate Thought-Action-Observation steps. We present Trajel, a dataset and evaluation framework for auditing trajectory-level hallucinations in multi-agent industrial workflows. Trajel introduces a five-type hallucination taxonomy (factual, referential, logical, procedural, and scope-based) over expert-annotated agent traces from AssetOpsBench. We benchmark supervised detection models at the subtask, trajectory, and long-context levels. Our results show that the most common failure modes are missed by existing benchmarks, that nearly half of hallucinated trajectories involve multiple types at once, and that automated detectors with high binary accuracy still misclassify the subtlest types. Trajectory-aware detection significantly outperforms standard post-hoc verification, making taxonomy-grounded evaluation necessary for safer agentic deployment.

Harshada Badave, Santosh Borse, Andrea Gomez +6

1IBM · 2Columbia University

AI Agent Reliability AI Agent Auditing LLM Auditing Hallucination in Language Models LLM Agent Evaluation AI Agent Benchmarks+2

May 22, 2026cs.LG

PromptAudit: Auditing Prompt Sensitivity in LLM-Based Vulnerability Detection

Large language models are increasingly used for vulnerability detection, yet their reliability under different prompt formulations remains uncharacterized. We present PromptAudit, a controlled evaluation framework that isolates prompt effects by fixing the dataset, decoding, and parsing while varying only the prompting strategy. Using five prompting strategies across five open-weight models on 1,000 CVEs (6,074 code samples spanning 16 programming languages), we evaluate accuracy, recall, abstention, coverage, and effective F1. We find that standard chain-of-thought prompting achieves the strongest overall operational performance, while few-shot prompting provides model-dependent benefits that are most pronounced for prompt-sensitive models. In contrast, adaptive chain-of-thought frequently suppresses recall and self-consistency induces excessive abstention, sharply reducing effective performance. These results show that vulnerability detection behavior is jointly determined by the model and the prompt, and that prompt sensitivity is a first-class system property that must be explicitly characterized in evaluation and deployment.

Steffen J. Camarato, Yahya Hmaiti, Mandana Ghadamian +1

University of Central Florida Orlando, Florida, USA

Prompt Sensitivity LLM Auditing Software Vulnerability Detection Prompt Engineering

May 22, 2026cs.SE

TRACER: A Semantic-Aware Framework for Fine-Grained Contamination Detection in Code LLMs

Data contamination is a known threat to the reliability of model evaluation. However, it remains underexplored in code large language models (LLMs), where contamination often goes beyond exact duplication. We present TRACER, a semantic-aware framework for fine-grained code contamination detection. TRACER models contamination using three levels of semantic overlap - Functionally Identical, Nearly Identical, and Shared Logic - and detects them through a coarse-to-fine pipeline. We also introduce the first benchmark for fine-grained code contamination detection, spanning three widely used benchmarks and three representative post-training datasets. TRACER achieves strong and consistent performance across multiple LLM backbones, with GPT-5 reaching an F1 score of 0.91 in fine-grained detection. In the binary setting, TRACER attains an F1 of 0.92, outperforming existing methods by 42%-217%. We further conduct ablation studies and error analysis to assess the contributions of individual components in TRACER.

Yifeng Di, Xuliang Huang, Tianyi Zhang

Purdue University West Lafayette, IN · University of Illinois Urbana-Champaign Champaign, IL

LLM Auditing Data Contamination in Language Models Code Language Models

May 22, 2026cs.AI

MemAudit: Post-hoc Auditing of Poisoned Agent Memory via Causal Attribution and Structural Anomaly Detection

Large language model agents increasingly rely on persistent memory to store past interactions, retrieve relevant demonstrations, and improve long-horizon task execution. However, this memory mechanism also creates a practical security vulnerability: an adversarial user may inject malicious records into the agent's memory through ordinary interaction, and these records can later be retrieved to steer the agent's reasoning and actions. Existing defenses primarily focus on online intervention, such as prompt filtering or output blocking, but they do not address the post-hoc question of which stored memories are responsible after harmful behavior has already been observed. We propose \textbf{MemAudit}, a post-hoc causal memory auditing framework for memory-augmented LLM agents. The framework combines two complementary signals: (1) a counterfactual memory influence score that measures each memory's causal contribution to harmful outputs, and (2) a memory consistency graph that identifies structurally anomalous memories within the broader memory store. We evaluate MemAudit against MINJA, a query-only memory injection attack in which malicious records are generated and stored through normal agent interactions rather than direct memory-bank modification. Across both QA and reasoning-agent settings, MemAudit substantially reduces attack success rates under realistic post-hoc auditing scenarios. The results show that QA attack success is reduced from

70\%

to

0\%

, while RAP attack success drops from

83.3\%

to

0\%

.

Zhewen Tan, Yilun Yao, Huiyan Jin +9

AI Agent Auditing LLM Auditing Causal Attribution Agent Memory Poisoning Memory-Augmented Language Model Agents

May 21, 2026cs.LG

Check Your LLM's Secret Dictionary! Five Lines of Code Reveal What Your LLM Learned (Including What It Shouldn't Have)

We show that singular value decomposition of the lm_head} weight matrix of a transformer-based large language model -- requiring only five lines of PyTorch and no model inference -- reveals interpretable semantic subspaces directly from the model weights. Each left singular vector identifies the vocabulary tokens most readily selected when the hidden state aligns with the corresponding singular direction; inspecting these clusters exposes the model's training data composition and curation philosophy. Analysing GPT-OSS-120B, Gemma-2-2B, and Qwen2.5-1.5B, we find that singular value spectra and vocabulary cluster structures differ systematically across models: GPT exhibits a graduated hierarchy of functionally differentiated subspaces; Gemma is dominated by pre-nineteenth-century English orthography, forming a stepwise clustering structure that may contribute to high output controllability; and Qwen exhibits broad multilingual coverage alongside subspaces whose vocabulary the authors have determined to be ethically inappropriate for direct publication. Base-instruct comparison reveals that ethically concerning subspaces originate in pretraining and are not removed by post-training alignment. We introduce the Vocabulary Cluster Score (VCS) to quantify subspace coherence, and the Weighted Projection Score (WPS) as a static glitch token detector; applying WPS to GPT-OSS-120B recovers shokubutsu-hyakka-tsu (ID 137606), a well-known glitch token widely reported in the CJK language community, without any model inference. We propose a taxonomy of root causes for problematic vocabulary content and call for lm_head} SVD analysis to be adopted as a standard pre-release safety auditing step. Our findings further suggest directions toward SVD-guided tokenizer optimisation and more controllable LLM design.

Hisashi Miyashita

Mgnite Inc.

LLM Auditing LLM Interpretability Language Model Probing

May 21, 2026cs.LG

The Illusion of Reasoning: Exposing Evasive Data Contamination in LLMs via Zero-CoT Truncation

Large language models (LLMs) have demonstrated impressive reasoning abilities across a wide range of tasks, but data contamination undermines the objective evaluation of these capabilities. This problem is further exacerbated by malicious model publishers who use evasive, or indirect, contamination strategies, such as paraphrasing benchmark data to evade existing detection methods and artificially boost leaderboard performance. Current approaches struggle to reliably detect such stealthy contamination. In this work, we uncover a critical phenomenon: a model's generated reasoning steps actively mask its underlying memorization. Inspired by this, we propose the Zero-CoT Probe (ZCP), a novel black-box detection method that deliberately truncates the entire Chain-of-Thought (CoT) process to expose latent shortcut mappings. To further isolate memorization from the model's intrinsic problem-solving capabilities, ZCP compares the model's zero-CoT performance on the original benchmark against an isomorphically perturbed reference dataset. Furthermore, we introduce Contamination Confidence, a metric that quantifies both the likelihood and severity of contamination, moving beyond simple binary classifications. Extensive experiments on both previously identified contaminated models and specially fine-tuned contaminated models demonstrate that ZCP robustly detects both direct and evasive data contamination. The code for ZCP is accessible at https://github.com/Yifan-Lan/zero-cot-probe.

Yifan Lan, Yuanpu Cao, Hanyu Wang +2

The Pennsylvania State University

LLM Evaluation LLM Auditing Benchmark Contamination Data Contamination in Language Models

May 20, 2026cs.CL

How Far Will They Go? Red-Teaming Online Influence with Large Language Models

As large language model (LLM)-based agents increasingly participate in online discourse, red-teaming their capacity to support political influence campaigns is critical for information integrity. In pursuit of this goal, we focus on locally deployed open-source LLMs, as opposed to frontier API-only models, given their superior alignment with the operational constraints of privacy-conscious malicious actors deployed in social media environments. We introduce an empirical red-teaming framework for measuring LLM Overton Windows (OWs), defined as the range of political opinions a model can reliably express on controversial topics, and for quantifying how simple natural-language jailbreaks expand that range. We evaluate more than 30 LLMs spanning 10 model families and five countries of origin. We find systematic asymmetries in political expressivity: open-source LLMs are typically more willing to generate left-leaning social media content, OWs tend to contract inversely to model size, and regional differences are substantial despite uneven representation in the open-source ecosystem. Jailbreak potency also varies sharply across model families, motivating a workflow for identifying effective combinations of jailbreak techniques. Taken together, our results establish a practical framework for auditing the political steerability of open-source LLMs and for helping future researchers design stronger countermeasures against LLM-enabled influence campaigns.

Daniel C. Ruiz, Anna Serbina, Ashwin Rao +2

Information Sciences Institute University of Southern California Los Angeles, CA, USA

Language Model Steering LLM Auditing Adversarial Attacks on LLMs Political Bias in Language Models LLM Red Teaming

May 20, 2026cs.LG

What Twelve LLM Agent Benchmark Papers Disclose About Themselves: A Pilot Audit and an Open Scoring Schema

We read twelve well-known LLM agent benchmark papers and recorded, dimension by dimension, what each paper actually says about how its evaluation was run. The motivation came from a familiar frustration: two papers will report results on the same benchmark with the same model name and disagree, and you cannot tell why -- the scaffold, the sampling settings, the subset, or the evaluator version. In many cases the published artifact does not let you answer. This paper is an implementation report on the attempt. We designed a small audit schema (five fields: benchmark identity, harness specification, inference settings, cost reporting, failure breakdown), wrote a scoring codebook with the boundary cases we hit during pilot scoring, applied it to twelve canonical papers (eight agent, four classical static), and recorded what we saw. We score the disclosure of an agent run, not its correctness, and make no claim that disclosure implies a trustworthy result. The mean audit score across the eight agent-benchmark papers is 0.38 (out of 1.0), and across the four classical static benchmarks 0.66; the largest gap is on cost (none of the eight agent benchmark papers disclose inference cost in any form) and on harness specification (none fully disclose a content-addressed container image of the evaluation environment). We release the schema as a JSON Schema file, the codebook as a Markdown document, and the raw scoring sheet as a CSV. The scoring was performed by a single auditor in one pass; a multi-rater audit is the natural next step, and we discuss what we think it would change.

Mahdi Naser Moghadasi, Faezeh Ghaderi

BrightMind AI, Texas Tech University · Research Division, BrightMind AI, Seattle, WA · Texas Tech University, Lubbock, TX +2

Benchmark Auditing LLM Auditing LLM Agent Evaluation AI Agent Benchmarks Scientific Reproducibility

May 20, 2026cs.CL

Do No Harm? Hallucination and Actor-Level Abuse in Web-Deployed Medical Large Language Models

Medical large language models (LLMs), including custom medical GPTs (MedGPTs) and open-source models, are increasingly deployed on web platforms to provide clinical guidance. However, they pose risks of hallucination, policy noncompliance, and unsafe design. We conduct a large-scale assessment of 6,233 MedGPTs, evaluating a stratified sample of 1,500, together with 10 open-source LLMs. We introduce two frameworks: MedGPT-HEval for hallucination detection and an LLM-based pipeline for assessing policy violations and developer intent. Our results show that 25-30% of MedGPTs exhibit low factual accuracy, with bottom- and middle-tier models at highest risk; 33.6-54.3% violate operational thresholds, and 57.06% of Action-enabled models lack adequate privacy disclosures. Compared with open-source models, MedGPTs achieve higher factual accuracy and semantic alignment, though open-source models are more stable. These results reveal systemic gaps in hallucination and compliance, highlighting the need for multi-metric evaluation and stronger safeguards. We release HAA-MedGPT, a structured dataset that supports future research on the safety of web-facing medical LLMs.

Sunday Oyinlola Ogundoyin, Muhammad Ikram, Rahat Masood

Macquarie University, Sydney, Australia · The University of New South Wales, Sydney, Australia

Language Model Safety Evaluation Healthcare LLM Auditing Hallucination in Language Models LLM Reliability LLM Hallucination Detection+2

May 19, 2026cs.CL

Stage-Audit: Auditable Source-Frontier Discovery for Cross-Wiki Tables

LLM-curated tables can appear source-grounded while containing unsupported rows: the curator may recall entries from parametric memory and retroactively attach page-level citations that are not the actual source. We study this hazard in Seed2Frontier discovery: the task of finding complement Wikipedia pages from a seed page to assemble a structured table. Stage-Audit addresses it with disjoint curator-auditor write rights, a row-level source-citation gate, and a 12-check audit taxonomy over keys, schema, source roles, cardinality, and scope. On a curated 51-instance Seed2Frontier evaluation set spanning 15 top-level domains, Stage-Audit improves source-frontier precision over a vanilla LLM curator from 0.356 to 0.505 (+42% relative) and F1 from 0.334 to 0.451 (+35%), while maintaining explicit per-row source traceability. The vanilla-LLM-vs-Stage-Audit comparison isolates the policy contribution rather than LLM-based discovery in general.

Chen Shen

Megagon Labs

Data Provenance LLM Auditing Information Retrieval

May 19, 2026cs.LG

The Evaluation Game: Beyond Static LLM Benchmarking

As jailbreaks, adversarially crafted inputs that bypass safety constraints, continue to be discovered in Large Language Models, practitioners increasingly rely on fine-tuning as a defensive strategy. Yet the theoretical foundations underlying this robustness fine-tuning remain underexplored. We introduce a game-theoretic framework in which the interaction between an evaluator (auditing the model for jailbreaks) and a trainer is formalized as a two-player game. A key feature of our approach is the use of group actions, a mathematical structure that captures symmetries and transformations, to formally represent data augmentation. The simplest non-trivial instance is the circle with cyclic translation groups, where we exhibit various regimes depending on the trainer's generalization range. Below a critical threshold, the evaluator maintains a constant miss ratio for linearly many rounds, whereas other settings can yield very different behaviors. We further provide empirical evidence supporting locality-dependence of the model: for the three model families we tested (Llama, Qwen and Mistral), we have significant evidence that fine-tuning on adversarial prompts induces only local generalization, with refusal rates on test examples highly correlated with the distance to the fine-tuning prompts. Our framework recasts the central object of adversarial evaluation: a benchmark is not a static set of prompts but an orbit under the evaluator's group action, and audit protocols that ignore trainer-side adaptation cannot distinguish a genuine fix from a memorized patch.

Paul Wang, Jade Garcia-Bourrée, Anne-Marie Kermarrec +1

Sorbonne Université, CNRS, LIP6 · École Polytechnique Fédérale de Lausanne

Adversarial Training LLM Auditing Adversarial Robustness of Language Models

May 19, 2026cs.CL

DECOR: Auditing LLM Deception via Information Manipulation Theory

Large language models can deceive by subtly manipulating truthful information -- omitting key facts, shifting focus, or obscuring meaning -- making such behavior difficult to detect. Existing black-box methods rely on coarse-grained judgments, offering limited interpretability and failing to pinpoint which facts were distorted and how. We introduce DECOR, a multi-agent framework grounded in Information Manipulation Theory for fine-grained auditing of strategic deception in LLM responses. DECOR decomposes input contexts into atomic informational units and scores each unit against the response across four dimensions of manipulation, producing interpretable manipulation profiles that are aggregated into a global deception index. We comprehensively evaluate DECOR on both single-turn and multi-turn deception detection benchmarks spanning real-world domains, and show that DECOR achieves state-of-the-art performance on both, outperforming competitive baselines. The framework generalizes across 15 frontier models, and ablation studies confirm the contribution of each key design component. Our findings demonstrate that fine-grained, theory-grounded auditing of information manipulation offers an effective and interpretable path for LLM deception detection.

Linyue Cai, Samuel Yeh, Jwala Dhamala +2

Department of Computer Sciences, University of Wisconsin-Madison · Amazon

Language Model Safety Evaluation LLM Auditing Deception in Language Models