Language Model Auditing

Follow this topic Ask about this topic

Language model auditing examines whether a model’s behavior, internal signals, training history, and deployed tools match claims about its reliability, safety, fairness, and compliance. Auditors use controlled prompts and counterfactual tests, multi-turn and domain-specific benchmarks, black-box output or likelihood analysis, white-box activation probes, and execution logs or claim-level evidence. These methods are applied to settings such as jailbreaks, retrieval-augmented generation, agent workflows, demographic bias, data leakage, and finetuning. A central challenge is validity: proxy scores and automated judges can miss meaningful failures or produce false alarms, so audits need calibrated thresholds, matched controls, realistic settings, and evidence that can be independently checked.

Latest papers

160 papers16 papers in the last four weeks, down 43% on the four weeks before. 0.1% of all new papers.

Beyond Attack Success Rate: Temporal Logit Observability for LLM Safety Failures
May 28, 2026Junyoung Park, Sunghwan Park, Seongyong Ju +1LLM Safety Evaluation LLM Jailbreak Attacks
Chung-Ang University
An Empirical Audit of k-NAF Budget Accounting for Anchored Decoding
May 27, 2026J. VijayavallabhConstrained Decoding Language Model Auditing
Indian Institute of Technology Madras Chennai, India
Auditing Stance Asymmetry in Generative Explanations
May 27, 2026Jiarui HanLanguage Model Generation Evaluation Social Bias in Language Models
QUACK: Questioning, Understanding, and Auditing Communicated Knowledge in Multimodal Social Deduction Agents
May 26, 2026Ye Yuan, Rui Song, Weien Li +12Social Deduction Games LLM Agent Evaluation
McGill University · Mila - Quebec AI Institute · University of Cambridge +3
Reading the Finetuning Prior: Verbatim Content Recovery via Contrastive Decoding Diffing
May 25, 2026Michał Brzozowski, Zuzanna Dubanowska, Enrico Cassano +1Training Data Extraction Language Model Fingerprinting
Samsung AI Center, Warsaw, Poland · University of Turin, Italy · University of Warsaw, Poland
ContextEcho: A Benchmark for Persona Drift in Long Agentic-Coding Sessions
May 22, 2026Xianzhong Ding, Yangyang Yu, Changwei Liu +3Tool-Augmented Language Model Agents AI Coding Agents
Center for Advanced AI, Accenture · Argonne National Laboratory · Independent Researcher
Reading Task Failure Off the Activations: A Sparse-Feature Audit of GPT-2 Small on Indirect Object Identification
May 21, 2026Mahdi NasermoghadasiTransformer Interpretability Sparse Autoencoders
Research Division, BrightMind AI, Seattle, WA · Texas Tech University, Lubbock, TX
Amplifying, Not Learning: Fine-Tuned AI Text Detectors Amplify a Pretrained Direction
May 20, 2026Alexander SmirnovAI-Generated Text Detection Classification
University College London
Lost in Interpretation: The Plausibility-Faithfulness Trade-off in Cross-Lingual Explanations
May 19, 2026Somnath Banerjee, Pranav Jha, Rima Hazra +1Language Model Auditing Cross-Lingual Reasoning in Language Models
Indian Institute of Technology Kharagpur · TCG Crest · National University of Singapore
RSD: Moving Local Triangular Charts for Auditing Language-Model Hidden States
May 17, 2026Seungmin JinWord Sense Disambiguation Hidden-State Probing
HSE University
TEMPO: Temporal Enforcement via Mode-Separated Policy Optimization for Trustworthy LLM Backtesting
May 13, 2026Zeyu Zhang, Bradly C. StadieLLM Evaluation RL for Language Model Reasoning
Department of Statistic and Data Science Northwestern University
Targeted Tests for LLM Reasoning: An Audit-Constrained Protocol
May 12, 2026Hongmin LiPrompt Sensitivity Language Model Robustness
School of Life Science and Technology, Institute of Science Tokyo · Department of Computational Biology and Medical Sciences, Graduate School of Frontier Sciences, The University of Tokyo
Benchmarking LLM-Based Static Analysis for Secure Smart Contract Development: Reliability, Limitations, and Potential Hybrid Solutions
May 11, 2026Stefan-Claudiu Susan, Andrei Arusoaie, Dorel LucanuLLM Evaluation Software Vulnerability Detection
Department of Computer Science Alexandru Ioan Cuza University of Ias,i Ias,i, Romania
Can You Keep a Secret? Involuntary Information Leakage in Language Model Writing
May 11, 2026Ari Holtzman, Peter WestData Leakage Privacy Leakage in Language Models
University of Chicago · University of British Columbia
The Translation Tax Is Not a Scalar: A Counterfactual Audit of English-Source Cue Inheritance in Chinese Multilingual Benchmarks
May 8, 2026Zezheng Lin, Fengming Liu, Handi LiMultilingual Language Model Evaluation Language Model Auditing
The Compliance Gap: Why AI Systems Promise to Follow Process Instructions but Don't
May 3, 2026Kwan Soo ShinAI Agent Benchmarks LLM Honesty
PolymathMinds AI Lab
SCALE-LoRA: Auditing Post-Retrieval LoRA Composition with Residual Merging and View Reliability
May 2, 2026Shuaipeng Zhou, Yu ZhangUncertainty Quantification Adapter Tuning
School of Computer Science and Technology, Shenyang University of Chemical Technology · 2Liaoning Key Laboratory of Intelligent Technology for Chemical Process Industry
Most Current Model Organisms Are Leaky: Perplexity Differencing Often Reveals Finetuning Objectives
May 1, 2026Mohammed Abu Baker, Luca Baroni, Dan WilhelmLanguage Model Safety Evaluation Language Model Auditing
Meridian Visiting Researcher Programme.
Semia: Auditing Agent Skills via Constraint-Guided Representation Synthesis
May 1, 2026Hongbo Wen, Ying Li, Hanzhi Liu +4AI Agent Auditing Static Code Analysis
University of California, Santa Barbara Santa Barbara, CA, USA · University of California, Los Angeles Los Angeles, CA, USA · Fuzzland USA +1
Latent Adversarial Detection: Adaptive Probing of LLM Activations for Multi-Turn Attack Detection
Apr 30, 2026Prashant KulkarniPrompt Injection Attacks on LLMs Adversarial Example Detection
Mountain View, CA
Political Bias Audits of LLMs Capture Sycophancy to the Inferred Auditor
Apr 30, 2026Petter Törnberg, Michelle SchimmelLLM Sycophancy Political Bias in Language Models
Institute of Logic, Language and Computation (ILLC), University of Amsterdam, Amsterdam, Netherlands
Evaluating Epistemic Guardrails in AI Reading Assistants: A Behavioral Audit of a Minimal Prototype
Apr 30, 2026Matthew Christian AgustinLanguage Model Safety Evaluation LLM Guardrails
Responsible Innovation Lab Tempe, Arizona, USA
Structured Security Auditing and Robustness Enhancement for Untrusted Agent Skills
Apr 28, 2026Lijia Lv, Xuehai Tang, Jie Wen +2AI Agent Auditing AI Agent Security
Institute of Information Engineering, Chinese Academy of Sciences · School of Cyber Security, University of Chinese Academy of Sciences
Architecture Determines Observability of Transformers
Apr 27, 2026Thomas CarmichaelTransformer Interpretability Language Model Error Detection
Independent Researcher
Escaping the Agreement Trap: Defensibility Signals for Evaluating Rule-Governed AI
Apr 22, 2026Michael O'Herlihy, Rosa CatalàContent Moderation Automated Evaluation
Reddit, Inc.
What Makes a Good AI Review? Concern-Level Diagnostics for AI Peer Review
Apr 21, 2026Ming JinAutomated Peer Review Generative AI Evaluation
Bradley Department of Electrical & Computer Engineering Virginia Tech
Semantic Needles in Document Haystacks: Sensitivity Testing of LLM-as-a-Judge Similarity Scoring
Apr 20, 2026Sinan G. Aksoy, Alexandra A. Sabrio, Erik VonKaenel +1Semantic Textual Similarity LLM-as-a-Judge
Pacific Northwest National Laboratory · Washington University in St. Louis · Humana Inc.
An Empirical Study of Multi-Generation Sampling for Jailbreak Detection in Large Language Models
Apr 20, 2026Hanrui Luo, Shreyank N GowdaLanguage Model Safety Evaluation LLM Auditing
School of Computer Science, University of Nottingham, Nottingham, United Kingdom.
Catching The Correct Answer Trap: Characterising AI Tutor Blind Spots When Analysing Student Reasoning
Apr 20, 2026Moiz Imran, Sahan BulathwelaIntelligent Tutoring Systems Educational Assessment
Department of Computer Science, University College London, UK · Centre for Artificial Intelligence, University College London, UK
The Geometric Canary: Predicting Steerability and Detecting Drift via Representational Stability
Apr 20, 2026Prashant C. RajuLanguage Model Steering LLM Reliability

May 28, 2026J/K move · Enter open · S save

Beyond Attack Success Rate: Temporal Logit Observability for LLM Safety Failures

Junyoung Park, Sunghwan Park, Seongyong Ju, Jaewoo Lee

Chung-Ang University

LLM Safety Evaluation LLM Jailbreak Attacks Language Model Auditing LLM Refusal Behavior Jailbreak Detection Refusal Behavior in Language Models+1

Open paper AskPDF arXiv

Attack Success Rate (ASR) evaluates each jailbreak with a single yes/no label at the end of generation, telling us whether a failure happened but not how it unfolded. Two attacks that produce equally harmful outputs may have followed completely different paths, and ASR cannot tell them apart. We make those hidden paths observable from logits alone. Temporal Logit Observability (TLO) is a training-free diagnostic that watches a compliance-refusal margin during decoding and places each model-attack condition on a calibrated 2D plane. By design, this plane is most informative exactly where ASR is least informative: among attacks that succeed for genuinely different reasons. Across four aligned LLMs and three jailbreak paradigms, attacks with nearly identical ASR land at clearly different points on the plane: the same model can fail through different temporal patterns. The geometry matches refusal-direction probes from hidden states on most conditions, with one model showing the limit of our fixed-lexicon approach. A simple early-stop rule derived from TLO cuts successful jailbreaks by more than half, without false alarms on plain benign queries. Safety evaluation should report when and how a failure unfolds, not only whether it occurred. TLO makes the first two observable from logits alone.