Skip to yearly menu bar Skip to main content

Workshop Fri, Oct 9, 2026 • 8:30 AM – 6:00 PM PDT Continental Ballroom 7,8

Actionable Interpretability

Tal haklay ⋅ Sarah Wiegreffe ⋅ Hadas Orgad ⋅ Mor Geva ⋅ Ian Tenney ⋅ Asaf Avrahamy ⋅ Anja Reusch ⋅ Marius Mosbach

Project Page

Abstract

The workshop on Actionable Interpretability@COLM2026 aims to foster discussions on leveraging interpretability insights to drive tangible advancements in AI across diverse domains. We welcome contributions that move beyond theoretical analysis, demonstrating concrete improvements in model alignment, robustness, and real-world applications. Additionally, we seek to explore the challenges inherent in translating interpretability research into actionable impact.

Chat is not available.

Schedule

Time zone: Conference local time zone

9:00 AM

Opening Remarks

9:10 AM

Keynote Tom McGrath, Goodfire - Ambitious Actionable Interpretability

Tom McGrath

9:50 AM

Keynote Dhanya Sridhar, Université de Montréal, Mila - Robust Interpretability with Causal Representation Learning

Dhanya Sridhar

10:30 AM

Shifting Mechanisms: How Positional Encoding Choice Shapes Long-Context Retrieval

10:40 AM

Self-CTRL: Self-Consistency Training with Reinforcement Learning

10:50 AM

Poster Session 1

11:55 AM

Lunch Break

1:30 PM

Keynote Yonatan Belinkov, Technion - (Actionable) Interpretability beyond Language

Yonatan Belinkov

2:15 PM

Verbalizing LLMs’ assumptions to explain and control sycophancy - a realistic scenario, connecting analysis with interpretability and acting on it

2:25 PM

Poster Session 2

3:30 PM

Coffee Break

4:00 PM

Keynote Christopher Potts, Stanford - Post Hoc, Ergo Propter Data

Christopher Potts

4:40 PM

Panel

5:10 PM

Closing Remarks

From Refusal to Fabrication: An Affective Cause of Hallucination That Bypasses the Honesty Gate

Akshat Agarwal ⋅ Brayden Feng ⋅ William Qian ⋅ Ruizhe Li ⋅ Aditya Shah ⋅ Vasu Sharma ⋅ Maheep Chaudhary

Fuzzing Large Language Models to Elicit Hidden Behaviours

Mohammed Abu Baker ⋅ Lakshmi Babu Saheer

Shifting Mechanisms: How Positional Encoding Choice Shapes Long-Context Retrieval

Eric Enouen ⋅ sainyam galhotra

MoRFI: Monotonic Sparse Autoencoder Feature Identification

Dimitris Dimakopoulos ⋅ Shay B Cohen ⋅ Ioannis Konstas

Who Is Left of Whom? Tracing Spatial Evidence and Role Binding in Relative-Position Reasoning

Yingjin Song ⋅ Denis Paperno ⋅ Albert Gatt

Confidence as a Compass: Regime-Dependent Dual Steering for Safety–Utility Tradeoffs in Language Models

Saleena Angeline Sartawita ⋅ Vanessa Matvei ⋅ Calvin Supasanya ⋅ Diyorbek Urozboev ⋅ Nyida Gyal ⋅ Shivank Garg ⋅ Kevin Zhu

When Is a Probe Monitor Safe to Act On? A Transferability Stress Test for Eval-Awareness Probes under RLVR

Sana Gaya ⋅ Imran Nasrullah ⋅ Cade Justad-Sandberg ⋅ Girish Gupta

Generalization of Steering Vectors for Chain-of-Thought Faithfulness

Matthew Nguyen ⋅ Kyle Cox ⋅ Austin Meek ⋅ Iván Arcuschin

Generative Meta-Models are Adversarial Classifiers

Diego Calanzone ⋅ Pietro Greiner ⋅ Pierre-Luc Bacon

Asymmetric Collapse in Model Merging: When Refusal Overwrites Recognition

Matheus Fonseca Rocha ⋅ Jiwon Seo ⋅ Aarnav Choudhary ⋅ Vasu Sharma ⋅ Maheep Chaudhary

LOCUS: Head Selection and Subspace Projection for Targeted Activation Steering.

Irene Tallini ⋅ Lorenzo Basile ⋅ Valentino Maiorca ⋅ Francesco Locatello ⋅ Alberto Cazzaniga

Truco-Bench: Covert Signaling in LLM Agents and What Black-Box Monitors Miss

Felipe Parodi

Look Before You Steer: Geometry Predicts SAE Feature Steerability

Muhammad Khan ⋅ Shlok Channawar ⋅ Akshaj Gurugubelli ⋅ Girish Gupta ⋅ Aditya Shah

Representation Steering for Emotions in Diffusion Language Models

Anil Ramakrishna ⋅ Shengyuan Hu ⋅ Will Bullock

Decodability is Not Causality: Dissociating Probe Readouts from Behavioral Drivers via SAE Decomposition

Devesh Tiwari ⋅ Camille Davis ⋅ Shivank Sinha ⋅ Talia Weaver ⋅ Aditya Shah ⋅ Maheep Chaudhary

From Refusal Tokens to Refusal Control: Discovering and Steering Category-Specific Refusal Directions

Rishab Alagharu ⋅ Ishneet S Singh ⋅ Shaibi Shamsudeen ⋅ Zhen Wu ⋅ Ashwinee Panda

Facial and Vocal Emotion Elicit Partially Shared Emotion-Sensitive Neurons in Multimodal Foundation Models

Xiutian Zhao ⋅ Luqi Sun ⋅ Berrak Sisman

Limits of Activation-Direction Interventions for Emergent Misalignment Beyond Rank-1 LoRA

Jayden Ramirez ⋅ Santiago Gutierrez ⋅ Anishansu Pradhan ⋅ Hikaru Isayama ⋅ Tinuade Adeleke

Trait-space Monitoring for Emergent Misalignment During Supervised Finetuning

Huy Nghiem ⋅ Sy-Tuyen Ho ⋅ Sarah Wiegreffe ⋅ Hal Daumé III

When the Loop Closes Quietly: Mechanistic Interpretability as Contestability Infrastructure for Deployed AI

Allan Ochola

Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration

Aashiq Muhamed ⋅ Mona Diab ⋅ Virginia Smith

Why Does Robustness Reduce Superposition?

ADAM ELIMADI

Localizing and Repairing Context-Failures in Audio-Language Models

Niclas Pokel ⋅ Pehuen Moure ⋅ Bilal Bounajma ⋅ Longbiao Cheng ⋅ Yingqiang Gao ⋅ Shih-Chii Liu ⋅ Roman Boehringer

Improving Confidence Estimation in Language Models with Linear Representations of Ambiguity

Themistoklis Nikas ⋅ Nishant Subramani ⋅ Aaron Mueller

Self-CTRL: Self-Consistency Training with Reinforcement Learning

Itamar Pres ⋅ Laura Ruis ⋅ Melat Ghebreselassie ⋅ Belinda Z Li ⋅ Jacob Andreas

Readable but Not Controllable: Neuron-Level Evidence for Medical LLM Hallucination

Vijay Vankadaru ⋅ Asha R Matthews ⋅ Tanya G Roosta ⋅ Peyman Passban

From Saliency to Actionability: Layer Localization Is Protocol-Conditional

Behnood Mashayekhi ⋅ Arian Amiramjadi ⋅ Parsa hosseini

Actionable Activation Directions for Detecting and Mitigating Emergent Misalignment Across Language Model Families

Abdul Rafay Syed

How Faithful Is Trajectory-Based Data Attribution? Error Sources, Remedies, and Practical Guideline

Junwei Deng ⋅ Pingbang Hu ⋅ Suliang Jin ⋅ Hao Lu ⋅ Jiachen T. Wang ⋅ Shichang Zhang ⋅ Jiaqi Ma

Guardrail Vectors: Bidirectional Steering and Framing-Induced Safety Failures

Paul Tylkin

LangFIR: Discovering Sparse Language-Specific Features from Monolingual Data for Language Steering

Sing Hieng Wong ⋅ Hassan Sajjad ⋅ A.B. Siddique

Decided Upstream, Written Late: Locating and Pricing the Cross-lingual Circuit of a Multilingual MoE

Ramakrishna P Kompella ⋅ Aadit Mahajan

RL-IF: Tracking Non-local Training Data Influence for Online Reinforcement Learning

Shixuan Liu ⋅ Cheng Tang ⋅ Yuzheng Hu ⋅ Fan Wu ⋅ Han Zhao ⋅ Jiaqi Ma

Targeting the Attention Heads Behind Object Hallucination in LLaVA

Armaan Sandhu ⋅ Abhilasha Senapati ⋅ Hima Jagadeesha Kammachi

Minimal Execution Patches for Causal Debugging of Tool-Using Agents

Tarika Gupta

Interpreting Prefill Awareness

Andrew Fletcher ⋅ Rohan Gupta

When Attention Closes: How LLMs Lose the Thread in Multi-Turn Interaction

Vardhan Dongre ⋅ Joseph Hsieh ⋅ Viet Lai ⋅ Seunghyun Yoon ⋅ Trung Bui ⋅ Dilek Hakkani-Tür

Simple Activation Clipping Defeats Circuit Breakers

George Yakushev ⋅ Kirill Pupkov

Sycophantic Agreement Transfers with Neutral Data via Contrastive Preference Optimization

Camila Blank ⋅ Zhuofan Ying ⋅ Christopher Potts ⋅ Peter Hase ⋅ Jing Huang

Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model

Pankayaraj Pathmanathan ⋅ Furong Huang

When Are Routing Weights Actionable? Causal Probes for Interpreting Block Attention Residuals

Aydin Javadov

Mechanistic Analysis of Safety Disparities Across Vision and Text Modalities in Vision-Language Models

Imran Kutianawala ⋅ Sarthak Tayal ⋅ Melissa Yu ⋅ Ayda Sultan

Tuning Language Models by Mixture-of-Depths Ensemble

Haoyan Luo ⋅ Lucia Specia

Before You Act: Benchmarking Interpretability in Vision-Language Models

Dia'a AL-Dweikat ⋅ Abdullah AlDahoul ⋅ Oyedeji Sonuga ⋅ Artur Shyutts ⋅ Jiahang He ⋅ Kevin Zhu ⋅ Ruizhe Li ⋅ Anusha Mujumdar

LayerRoPE: Dynamic Depth-wise Magnitude & Angular Superposition

Shikhar Srivastava ⋅ Christopher Kanan

RARE: Decoupling Representation Steering from Expert Routing in Mixture-of-Experts Language Models

Zhibo Zhang ⋅ Zhen Ouyang ⋅ Ling Shi ⋅ Kailong Wang

Spectral Query-Key Product Weight Steering for Training-Free VLM Hallucination Mitigation

Karn Tiwari ⋅ Varnith Chordia ⋅ Prathosh AP

Mechanistic Specialization Does Not Guarantee Performance: Why GPT-2 Outperforms Dual Attention Transformers on Identity Rule Tasks

Jonathan Scott ⋅ Leilani H Gilpin ⋅ Adrian Brasoveanu

CircuitSteer: Geometrically Aligned Multi-Layer Steering via Sparse Autoencoder Circuits

Mehrshad Saadatinia ⋅ Parsa Razmara ⋅ Ardalan Aryashad ⋅ Ali Abbasi ⋅ Seyedarmin Azizi

Characterizing Model-Native Skills

Feiyang Kang ⋅ Mahavir Dabas ⋅ Myeongseob Ko ⋅ Ruoxi Jia

Don’t Blink: Evidence Collapse during Multimodal Reasoning

Suresh Raghu ⋅ Satwik Pandey

EMO: Pretraining Mixture of Experts for Emergent Modularity

Ryan Wang ⋅ Akshita Bhagia ⋅ Sewon Min

On the Non-Identifiability of Steering Vectors in Large Language Models

Sohan Venkatesh ⋅ Ashish M Kurapath