cs.CLNov 16, 2026

Abstention Direction in Large Language Models: Mechanistic Interpretability of Abstention Across Query Ambiguity, Knowledge Limits, and Safety Alignment

Authors: Koki Itai, Jun Hirako, Hiroaki Yamagiwa

Organizations: SB Intuitions Corp.

Abstract

Abstention, the withholding of a response by a large language model (LLM) under uncertainty or safety risks, is essential for trustworthy models. The causes of abstention fall into three perspectives: ambiguity in the Query (Q), limits of Model Knowledge (MK), and misalignment with Human Values (HV). However, prior analyses of internal representations have focused on HV, using a refusal direction computed from harmful and harmless queries. No study has analyzed the three perspectives jointly. We therefore extend the refusal direction framework to all three perspectives. Specifically, for each perspective, we construct counterfactual pairs that share a template and differ only in the abstention-inducing factor, and compute an abstention direction by normalizing the difference in mean activations between the two polarities. Activation steering experiments show that intervening in middle layers increases or decreases abstention across all three perspectives. Moreover, steering yields cross-perspective effects, indicating that the abstention directions share a common directional component.

Figures & tables

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Jul 22, 2025

LLM Abstention Can Be a Prompt Artifact, in Addition to Genuine Uncertainty

Large Language Models (LLMs) are increasingly trained to abstain from answering questions they are unsure about. However, this ability is often misapplied: in real-world applications, user prompts sometimes contain elements of uncertainty, which lead LLMs to abstain even on problems they are capable of solving. We argue that LLM abstention is not only an expression of genuine uncertainty; it can also be an artifact largely shaped by prompts. We name this phenomenon Abstention Inflation. We add "Unknown" as an extra option for LLMs to choose from; experiments show serious accuracy drops on True/False Questions (TFQs). Replacing "Unknown" with an unrelated random word produces a similar effect. We argue that LLMs are trained to imitate the surface pattern of abstention, rather than to express genuine uncertainty. Based on ten experimental settings, we support four claims that form a progressive argument: (C1) Abstention Inflation can be triggered by the presence of an extra option, not by genuine uncertainty; (C2) it makes the models deny they can answer, even when they can; (C3) it is a later-layer output override, as the reasoning traces and mid-layer representations preserve correct answers; (C4) it is not stochastic noise: it results from various factors, emerges through instruction tuning, is boosted by problems' higher difficulty, and can be mitigated at larger model sizes.
Apr 2, 2026

Answering the Wrong Question: Reasoning Trace Inversion for Abstention in LLMs

For Large Language Models (LLMs) to be reliably deployed, models must effectively know when not to answer: abstain. Reasoning models, in particular, have gained attention for impressive performance on complex tasks. However, reasoning models have been shown to have worse abstention abilities. Taking the vulnerabilities of reasoning models into account, we propose our Query Misalignment Framework. Hallucinations resulting in failed abstention can be reinterpreted as LLMs answering the wrong question (rather than answering a question incorrectly). Based on this framework, we develop a new class of state-of-the-art abstention methods called Trace Inversion. First, we generate the reasoning trace of a model. Based on only the trace, we then reconstruct the most likely query that the model responded to. Finally, we compare the initial query with the reconstructed query. Low similarity score between the initial query and reconstructed query suggests that the model likely answered the question incorrectly and is flagged to abstain. Extensive experiments demonstrate that Trace Inversion effectively boosts abstention performance in four frontier LLMs across nine abstention QA datasets, beating competitive baselines in 33 out of 36 settings.
Oct 12, 2026

Strong yet Controllable LLM Abstention without Learning to Abstain

Large language models (LLMs) can reduce hallucination by abstaining from questions they would likely get wrong. Abstention can be implemented outside the model (selective answering) using a confidence threshold that controls the answer rate (coverage), or learned inside it (abstention training), which can perform better but fixes coverage once trained. To compare the two, we match their coverage and decompose risk (error rate) into answering ability and distinguishing ability, i.e., how well the model answers and how accurately it abstains, respectively. The decomposition shows that abstention training performs better mainly by answering better: with strong confidence estimators, the two have nearly equal distinguishing ability. This motivates Train for Accuracy, then Post-hoc Select (TAPS), a modular approach that keeps coverage controllable, needs only standard post-training and an off-the-shelf confidence estimator, and benefits from advances in either component. Across three LLMs and three factual QA benchmarks, TAPS with a probing-based confidence estimator attains 1.5--9.4 percentage points lower mean risk than three representative abstention training methods at matched coverage. Within TAPS, accuracy training often weakens the model's own confidence signals, yet refitting the estimator largely preserves distinguishing ability. Together, these results establish TAPS as a strong yet controllable baseline and raise the bar for future abstention training: sacrificing control over coverage should yield gains beyond what can be achieved by optimizing answering and abstaining separately.