Pith. sign in

REVIEW 6 cited by

Understanding Refusal in Language Models with Sparse Autoencoders

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2505.23556 v1 pith:6UZGADZM submitted 2025-05-29 cs.CL

Understanding Refusal in Language Models with Sparse Autoencoders

classification cs.CL
keywords refusalfeaturesmodelsadversarialautoencoderslanguagelatentmechanisms
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Refusal is a key safety behavior in aligned language models, yet the internal mechanisms driving refusals remain opaque. In this work, we conduct a mechanistic study of refusal in instruction-tuned LLMs using sparse autoencoders to identify latent features that causally mediate refusal behaviors. We apply our method to two open-source chat models and intervene on refusal-related features to assess their influence on generation, validating their behavioral impact across multiple harmful datasets. This enables a fine-grained inspection of how refusal manifests at the activation level and addresses key research questions such as investigating upstream-downstream latent relationship and understanding the mechanisms of adversarial jailbreaking techniques. We also establish the usefulness of refusal features in enhancing generalization for linear probes to out-of-distribution adversarial samples in classification tasks. We open source our code in https://github.com/wj210/refusal_sae.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Do Activation Monitors Survive Model Updates? Benchmarking, Predicting, and Repairing Activation-Monitor Staleness

    cs.LG 2026-06 unverdicted novelty 8.0

    Fine-tuning updates frequently stale activation monitors for language model safety while quantization does not, with degradation predictable and repairable via label-free realignment.

  2. Do LLMs Know Their Vulnerable Scenarios?

    cs.AI 2026-07 conditional novelty 7.0

    Scenario jailbreaks suppress refusal via internal concept directions; Concept2Scenario attributes those concepts with SAEs and turns them into transferable natural-language attack scenarios.

  3. OpenSafeIntent: Evaluating Intent-Calibrated Safe Completion Across Dual-Use Prompt Sets

    cs.CL 2026-07 unverdicted novelty 7.0

    OpenSafeIntent benchmark shows models fail to calibrate safety across intent shifts in matched dual-use prompts, indicating current evaluations are insufficient.

  4. Minionese: Comprehensive Benchmark and Mechanistic Study of Multilingual LLM Safety

    cs.CR 2026-07 conditional novelty 6.5

    Multilingual jailbreaks succeed mainly by routing harm through misaligned residual subspaces that under-project onto an intact, effectively one-dimensional cross-lingual refusal direction.

  5. Faithfulness to Refusal: A Causal Audit of Neuron Selectors

    cs.CL 2026-07 conditional novelty 6.0

    A causal audit via neuron-row zeroing shows attribution methods (LRP, IG) faithfully identify dispensable neurons and can install refusal behavior, while rank-stability proxies systematically miss selector failures.

  6. Beyond "I cannot fulfill this request": Alleviating Rigid Rejection in LLMs via Label Enhancement

    cs.CL 2026-05 unverdicted novelty 5.0

    LANCE applies variational inference for label enhancement across multiple rejection categories, supplying gradients to a refinement model that produces safe, non-rigid responses from LLMs.