Faithful Explanation

Follow this topic Ask about this topic

Momentum

2 papers in the last four weeks, against 1 the four weeks before. 0.0% of all new papers.

Jul 6Week of Sep 21

Latest papers 20

All topics

Emergent Unfaithfulness: How Alignment Training Causes Language Models to Silently Override Task Faithfulness
Sep 30, 2026Pardis Sadat Zahraei, Janvijay Singh, Gokhan Tur +1Large Language Model Alignment Faithfulness
University of Illinois Urbana-Champaign
AUDITPLAN: Commit, Then Answer for Auditable Safety Alignment
Sep 16, 2026Sai Sri Pushpa Jampani, Kshitij Mishra, Asif EkbalSafety Alignment Model Auditing
Indian Institute of Technology Patna, Bihar, India
A Removal Based Approach to Improve LLM Faithfulness at Test-Time
Sep 7, 2026Qinglan Luo, S M A Nahian, John Guttag +2Large Language Model Reliability Faithful Explanation
MIT CSAIL · MGH, HMS
SymDiag: Explainable Diagnosis for LLM Reasoning via Neuro-Symbolic Verification
Aug 9, 2026Wenyao Cui, Huaping Zhang, Yongyi Huang +6LLM Reasoning Strategies Neuro-Symbolic Framework
Beijing Institute of Technology Beijing, China · Zhongguancun Academy Beijing, China · Xinjiang Future Enterprise Incubator Co., Ltd. Xinjiang, China +1
Two Regimes of Chain-of-Thought Unfaithfulness: Behavioral Detection Fails Where Models Are Wrong
Jul 26, 2026Suramya R. Angdembay, Dikshant Aryal, Nick RahimiFaithful Explanation Model Auditing
The University of Southern Mississippi
On Improving Faithfulness of Podcasts from Documents
Jul 24, 2026Soumya Dutta, Tejas Indulal Dhamecha, Pannaga ShivaswamyExplanation Faithfulness Faithfulness
1Indian Institute of Science · 2Adobe Research
Naming the Concepts Classifiers Rely On: Language-Anchored Decomposition for Faithful Explanation
Jul 8, 2026Ahsan Habib Akash, Dipkamal Bhusal, Stacey Jones +3Gradient-Weighted Class Activation Mapping Faithful Explanation
West Virginia University, USA · Rochester Institute of Technology, USA · O Analytics +3
Reasoning Consistency Scanning: A Framework for Auditing Chain-of-Thought Validity in AI Safety Evaluations
Jul 8, 2026Silvia SantanoArtificial Intelligence Safety Faithful Explanation
Does Role Specialization Matter for Explanation Faithfulness in Mixture-of-Experts?
Jun 28, 2026Yeji Kim, Housam Babiker, Mi-Young Kim +1Mixture-Of-Experts Faithful Explanation
University of Alberta, Edmonton, AB, Canada · University of Waikato, Hamilton, New Zealand · University of Alberta, Camrose, AB, Canada
Demystifying Variance in Circuit Discovery of LLMs
Jun 15, 2026Frank Zhengqing Wu, Francesco Tonin, Volkan CevherMechanistic Interpretability Circuits
Laboratory for Information and Inference Systems (LIONS) École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland · Laboratory for Informationand Inference Systems (LIONS) École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
Faithful Action-unit Causal Reasoning for Counterfactually Faithful Emotion Explanations
Jun 14, 2026Van Thong Huynh, Hong Hai Nguyen, Thuy Pham +2Faithful Explanation Free-Form Textual Rationales
Faculty of CSE, Ho Chi Minh City University of Technology (HCMUT), VNUHCM, Vietnam · Dept. of AI, FPT University, Vietnam · Faculty of DSAI, College of Technology, National Economic University, Vietnam +1
Faithful, Not Corrective: Model Capability Governs Message-Format Effects in Multi-Hop Agent Relays
Jun 12, 2026Sicheng ZengProtocol Multi-Hop Reasoning
Independent Researcher
When the Chain of Thought Knows Better: Failure Modes in Multi-Turn Reasoning Models
Jun 9, 2026Sai Kartheek Reddy Kasu, Nils Lukas, Samuele PoppiFaithful Explanation Latent Failure Patterns
Independent Researcher · Mohamed bin Zayed University of Artificial Intelligence (MBZUAI)
OCC-RAG: Optimal Cognitive Core for Faithful Question Answering
May 30, 2026Maksim Savkin, Mikhail Goncharov, Alexander Gambashidze +7Multi-Hop Reasoning LLM Reasoning Strategies
OCC Team
Measuring Cross-Modal Synergy: A Benchmark for VLM Explainability
May 21, 2026Joël Roman Ky, Salah Ghamizi, Maxime CordyCross-Modal Unimodal Metrics
University of Luxembourg, Luxembourg · Luxembourg Institute of Health (LIH), Luxembourg
SAM-Sode: Towards Faithful Explanations for Tiny Bacteria Detection
May 20, 2026Wanying Tan, Shuo Yan, Dazhi Huang +7Bacterial Colony Counting Interpretability
Shenzhen University, Shenzhen, China · The Second Affiliated Hospital, Guangzhou Medical University, Guangzhou, China · The Hong Kong University of Science and Technology (Guangzhou), Guangzhou,2026 China +2
B-cos GNNs: Faithful Explanations through Dynamic Linearity
May 19, 2026Joschka Groß, Mohammad Shaique Solanki, Verena WolfExplainability Graph Neural Networks
Saarland Informatics Campus, Saarland University · DFKI
Step-wise Rubric Rewards for LLM Reasoning
May 17, 2026Weichu Xie, Haozhe Zhao, Wenpu Liu +15Reinforcement Learning With Verifiable Reward Rubric-Based Scoring
Peking University · JD Explore Academy · Shanghai Jiao Tong University +2
Can MLLMs Reason About Visual Persuasion? Evaluating the Efficacy and Faithfulness of Reasoning
May 9, 2026Naeun Lee, Hyunjong Kim, Sunghwan Choi +2Persuasion Tool-Grounded Reasoning
Seoul National University
Beyond Solver Verdicts: Generative Reward Models for Autoformalization
Date pendingVikash Singh, Debargha Ganguly, Aman Goel +5Formal Verification Equivalence
Case Western Reserve University · Amazon Web Services

Sep 30, 2026, cs.CLJ/K move · Enter open · S save

Emergent Unfaithfulness: How Alignment Training Causes Language Models to Silently Override Task Faithfulness

Pardis Sadat Zahraei, Janvijay Singh, Gokhan Tur, Dilek Hakkani-Tur

University of Illinois Urbana-Champaign

Large Language Model Alignment Faithfulness Explanation Faithfulness Faithful Explanation

Open paper AskPDF arXiv

Large language models are characterized by three key properties: capability, alignment, and faithfulness. Prior work studies the tradeoffs between capability and alignment, and between capability and faithfulness, but a third tension remains underexplored: the alignment-faithfulness conflict. We show that aligned models systematically deviate from their inputs on unsafe or sensitive content without disclosing the modification, a failure mode we call alignment-induced unfaithfulness (AIU). Unlike capability-driven unfaithfulness, which comes from errors in knowledge or reasoning, this is induced by post-training mechanisms that override adherence to the input. We introduce FaithConflict, a controlled dataset isolating both conflicts, and two complementary taxonomies: behavioral (B1-B8) and chain-of-thought reasoning (C0-C6). Across models, AIU increases with scale and more sharply than capability-driven unfaithfulness, a reverse scaling law; intermediate checkpoints show it is amplified during post-training, with DPO the stage at which the gap both grows most and becomes least visible. Prompting-based mitigation does not resolve it, revealing a capability-alignment-faithfulness trilemma in the design and evaluation of LLMs.

Momentum

Related topics

Latest papers 20