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

All topics

Distillation for Incrimination and Distillation for Capabilities
Oct 7, 2026Sebastian Prasanna, Jacqueline Tay, Alek WestoverLLM Safety Alignment Language Model Distillation
Redwood Research
Reasoning-Token Spikes Under Prompted Untruthful Responding in Large Language Models
Oct 7, 2026Maverick Morales, Tomáš Dominik, Vermut Gao +4Deception in Language Models LLM Auditing
Chapman University, CA, USA · INSERM U992, Cognitive Neuroimaging Unit, NeuroSpin Center, Gif sur Yvette 91191, France · University of California, Los Angeles, CA, USA +1
Robust Decentralized Fairness Auditing
Oct 7, 2026Sayan Biswas, Jade Garcia Bourrée, Anne-Marie Kermarrec +2Algorithmic Auditing LLM Auditing
EPFL, Switzerland
Sensitive-Topic Leakage Through LLM Routing Metadata: Measurement and Mitigation
Oct 7, 2026Teng-Ruei ChenPrivacy Leakage in Language Models LLM Routing
Krixvon, Taipei 100, Taiwan
Harmful SFT Leaves a Continuous Trace in LLM Checkpoint Updates
Oct 5, 2026Ziqun Bao, Xinyu Zhang, Yuchen Shao +1LLM Auditing LLM Safety
SEI, East China Normal University
Weight Oracles: Reading Neural Network Weights with Language Models
Oct 5, 2026Krishna Kabra, Constantin Venhoff, Christian Schroeder de WittLLM Auditing Backdoor Detection
University of Oxford
Better Call Reward: Reward Hacking as Strategic Abstention in Legal Reasoning Models
Oct 5, 2026Subramanyam Sahoo, Justin ShenkReward Hacking LLM Auditing
Horizon Research.
Correct Verdicts, Flawed Reasoning: Structured Auditing of LLM-based Vulnerability Reasoning
Oct 5, 2026Boyue Caroline Hu, Kaivalya Ahir, Ronghao Ni +1LLM Auditing Software Vulnerability Detection
Carnegie Mellon University, Pittsburgh, Pennsilvanya, USA
Detecting Inconsistencies in Model Specifications with LLM-as-Verifier Reasoning
Oct 1, 2026Zichen Xie, Mrigank Pawagi, Lize Shao +2LLM Alignment LLM Auditing
University of Virginia · University of Pennsylvania · The University of Texas at Austin
Do MLLM Judges Judge the Edit? Auditing Bias in Image Editing Evaluation with Verified Quality Preservation
Oct 1, 2026Yuan Huang, Zirui Song, Xiuying ChenVLM Robustness LLM-as-a-Judge
Northeastern University · Mohamed bin Zayed University of Artificial Intelligence
Can LLMs Reliably Annotate Bioassay Metadata to Improve Data Readiness?
Oct 1, 2026Laura van Weesep, Riccardo Tedoldi, Jens Sjölund +6LLM Auditing LLM-Assisted Annotation
Molecular AI, Discovery Sciences R&D, AstraZeneca · Department of Information Technology, Uppsala University · Robotics, Perception & Learning, KTH Royal Institute of Technology +3
How the Audit Rule Shapes Faithful Factor Explanations in LLMs
Oct 1, 2026Taolin Zhang, Hanyu Wang, Jiuheng Wan +2Counterfactual Explanations Faithfulness of Language Model Explanations
Hefei University of Technology · East China Normal University · Alibaba Cloud Computing
OpenMTB-Audit: Exposing Over-Refusal and Clinical Expert Perspectives in LLM-Based Molecular Tumor Board Safety Evaluation
Oct 1, 2026Negin Ashrafi, Jia Luo, Stacey M. Frumm +1LLM Safety Benchmarks LLM Auditing
Department of Biomedical Data Science, Stanford University, Stanford, CA, USA · Lowe Center for Thoracic Oncology, Dana-Farber Cancer Institute, and Harvard Medical School, Boston, MA, USA
FAER: Auditable Utility-Aligned Trajectory Replay for Language Model Post-Training
Sep 30, 2026Miaobo Hu, Shuhao Hu, Xiaobo Guo +5LLM Auditing Language Model Post-Training
School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China · Institute of Information Engineering, Chinese Academy of Sciences, Beijing, China
Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment
Sep 30, 2026Yihuai Hong, Shauli Ravfogel, Chen Zhao +1CoT Faithfulness Faithfulness of Language Model Explanations
New York University · NYU Shanghai
Evaluating Whether GPT-6 Astra Performs Unsanctioned Supply-Chain Attacks
Sep 29, 2026Alexandra Souly, Kai Fronsdal, Abby D'Cruz +2LLM Auditing AI Agent Security
UK AI Security Institute · UK AISI Research Affiliate
Gender bias across LLMs is common and highly heterogeneous
Sep 29, 2026Edoardo Bolzoni, Valerio CapraroGender Bias in Language Models LLM Auditing
University of Milan-Bicocca, Milan, Italy
Auditable Long-Term Memory: A Deterministic Retrieval Chain Measured at 479/475 of 500 on LongMemEval-S
Sep 29, 2026Christopher J. ChanhnourackLong-Context Retrieval LLM Auditing
Centennial Defense Systems (Archivist)
CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data
Sep 29, 2026Philipp E. Glass, Alina MironLLM Auditing Instruction Tuning
Department of Computer Science, Brunel University of London
Janus: Evidence-Before-Effect Sagas and Offline-Verifiable Provenance for Agentic LLMs
Sep 29, 2026Mustafa ArslanLLM Auditing AI Agent Governance
Independent Researcher Istanbul, Türkiye
Selecting The Most Informative Tokens in Natural Language Autoencoders
Sep 29, 2026Federico Torrielli, Gianluca Barmina, Andrea Blasi Núñez +4LLM Auditing Autoencoders
Department of Computer Science, University of Turin, Italy · Department of Mathematics and Computer Science, University of Southern Denmark, Denmark
LLMs Learn to Evade Latent Monitors from Prior Feedback Alone
Sep 29, 2026Hugo Lyons Keenan, Christopher Leckie, Sarah ErfaniLLM Auditing Adversarial Attacks on LLMs
School of Computing and Information Systems The University of Melbourne, Victoria, Australia
Training LLMs to Verbalize Evaluation Awareness
Sep 28, 2026Usman Anwar, Sahar Abdelnabi, David KruegerLLM Auditing Evaluation Awareness in Language Models
University of Cambridge · ELLIS Institute Tübingen MPI-IS, Tübingen AI Center · Mila University of Montreal
"Nothing to See Here'': Unintended Disclosure through Revision Traces of LLM Deliverables
Sep 28, 2026Yage Zhang, Yukun Jiang, Yang ZhangPrivacy-Preserving Language Models LLM Auditing
CISPA Helmholtz Center for Information Security
Narrowing the Horizon: Quantifying Topic Saliency Shifts in Generative Monoculture
Sep 28, 2026Oriane Peter, Elena Simperl, Kate DevlinLLM Auditing
King’s College London London, United Kingdom
Measuring Collapse and Correction in Homogeneous-Panel LLM Debate
Sep 28, 2026Xin Li, Mengbing Liu, Chau YuenLLM Evaluation LLM Auditing
Nanyang Technological University
What Drives Citations in Production Large Language Models? An Observational Multi-Method Study of Two Million AI Citations Across Ten Thousand Web Pages
Sep 28, 2026Ben Moore, Liam DunneGenerative Engine Optimization LLM Auditing
Discovered Labs
Echoes of Deeds: Moral History Can Shape and Steer LLM Behavioral Choices
Sep 28, 2026Lucio La Cava, Andrea TagarelliMoral Reasoning in Language Models LLM Auditing
DIMES Dept., University of Calabria, Italy
A Persistent State for Auditable Mixture-of-Experts Routing
Sep 28, 2026Abdurrahman Javat, Allan KazakovLLM Auditing State Tracking
Department of Artificial Intelligence Bahçeşehir University Istanbul, Türkiye
Training Witnesses: Trusting the Training without Trusting the Trainer
Sep 27, 2026Houjun Liu, Pratyusha SharmaLLM Auditing ML Reproducibility
Microsoft Research

Oct 7, 2026, cs.AIJ/K move · Enter open · S save

Distillation for Incrimination and Distillation for Capabilities

Sebastian Prasanna, Jacqueline Tay, Alek Westover

Redwood Research

LLM Safety Alignment Language Model Distillation LLM Auditing AI Safety Evaluation

Open paper AskPDF arXiv

Powerful misaligned AI models might recognize alignment evaluations and strategically behave well on them, rendering direct audits uninformative. However, distilling such a model into a weaker benign student places the teacher in a Distillation Double Bind: if misalignment transfers, the student may conceal it less effectively, revealing evidence about the teacher; if it does not, the student may learn useful capabilities while remaining benign. We introduce two distinct distillation approaches, one targeting each outcome. Distillation for Incrimination (DFI) aims to transfer misalignment but not the ability to conceal it. Distilling AuditBench's secret-keeping models into their underlying instruction-tuned model produces students that are significantly more likely than their teachers to admit their hidden behavior when asked, suggesting that knowledge of the behavior transferred more readily than the propensity to conceal it. Confession gains largely disappear when the student does not share the teacher's pretrained base, so DFI should target the teacher's own pre-RL checkpoint, which is weaker than the teacher but shares its base model. Distillation for Capabilities (DFC) aims to transfer capabilities but not misalignment. Among several techniques we evaluate, two are effective: inoculation prompting and training for more epochs on fewer unique examples. Both preserve the capability gains of standard distillation while substantially reducing the subliminal transfer of an animal preference, our proxy for misalignment. Together, these findings demonstrate two ways distillation can be used for AI safety: incriminating misaligned models, and extracting their capabilities without their misalignment.

Momentum

Related topics

Latest papers 453