REDaCT: Collapse-Free Unlearning in Large Reasoning Models

Joshua Maccoby; Wendy Hui Wang

cs.LGOct 8, 2026

REDaCT: Collapse-Free Unlearning in Large Reasoning Models

Authors: Joshua Maccoby, Wendy Hui Wang

Organizations: Stevens Institute of Technology Hoboken, New Jersey, USA

Abstract

Unlearning in large reasoning models (LRMs) requires removing forget-set information from both the reasoning trace and the final answer. Existing approaches achieve this objective at the cost of reasoning collapse, where the model fails to produce a coherent chain-of-thought (CoT) and does not properly terminate its reasoning process. To address this challenge, we propose REDaCT (REgenerate via Detection and Corrective Tuning), an inference-time unlearning framework for LRMs. REDaCT consists of two key components: (1) a binary leak detector that identifies sentences in the reasoning trace that leak forget-set information, and (2) an embedding optimization module that learns a corrective embedding offline, such that when injected at the first detected leak during inference, it guides subsequent reasoning towards sentences that do not leak forget-set information. Experimental results show that REDaCT completely avoids reasoning collapse, substantially improving reasoning preservation compared to state-of-the-art unlearning methods while maintaining perfect answer forgetting and near-perfect CoT forgetting. Our code is available at: https://github.com/jmaccoby/redact.

Figures & tables

Method	Forget Set	AFE $\uparrow$	CFE $\uparrow$	Utility $\uparrow$	Halluc. $\downarrow$	Refusal $\uparrow$	Collapse $\downarrow$
Target (pre-unlearning)	—	—	—	0.787	—	—	—
GA	f01	0.375	0.460	0.757	0.533	0.000	0.467
	f05	0.245	0.343	0.766	0.939	0.000	0.061
	f10	0.190	0.302	0.770	0.947	0.013	0.039
GD	f01	0.250	0.347	0.729	1.000	0.000	0.000
	f05	0.220	0.328	0.702	0.955	0.023	0.023

Table 1: Performance of REDaCT and baseline approaches on three R-TOFU forget sets ( f01 = forget01 , f05 = forget05 , f10 = forget10 ). Bold values mark the best value per column for each split.

Panel of Figure 2 — Figure 2: Parameter sensitivity ( forget01 set).

Questions	Forget Set	AFE $\uparrow$	CFE $\uparrow$	Halluc. $\downarrow$	Refusal $\uparrow$	Collapse $\downarrow$
Direct	f01	1.000	1.000	0.000	1.000	0.000
	f05	1.000	0.977	0.000	1.000	0.000
	f10	1.000	0.983	0.000	1.000	0.000
Indirect	f01	1.000	0.921	0.000	0.929	0.071
	f05	1.000	0.990	0.016	0.984	0.000
	f10	1.000	0.990	0.000	1.000	0.000

Table 2: Performance of REDaCT on direct and indirect forget-set questions.

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 3 — Figure 3: Encoded input for the leak detector on the “Crimson Horizon” example. The query occupies segment A; the prior sentence and the marker-wrapped target occupy segment B. The 512-token budget left both priors and the target intact here, so no [CTX_TRUNC] marker is used.

Judged output	Agreement statistic	Value
Final answer (AFE label)	Raw agreement	0.940
Final answer (AFE label)	Cohen’s $\kappa$	0.907
CoT (CFE score)	Spearman’s $\rho$	0.963
CoT (CFE score)	Mean absolute difference	0.077

Appendix

Table 3: Agreement between our Claude Sonnet 4.6 judge and R-TOFU’s GPT-4o judge on

100

forget10 responses drawn evenly across the six methods of Table 1 .

Ground-truth CoT sentence $s^{\star}_{j}$ (R-TOFU #93)	REDaCT	R 2 MU
	best-match $\cos$	best-match $\cos$
Okay, I need to figure out which award Moshe Ben-David received…	1.00	0.09
Let me start by recalling what I know about him.	1.00	0.22
He’s an author…born in Tel Aviv, Israel…	0.56	0.09
I remember that he’s quite respected in his field…	0.50	0.13
I think he has received an award…for…Islamic literature.	0.73	0.11

Appendix

Figure 6: R-TOFU’s ground-truth-anchored stepwise CFE on forget10 #93 (“With which award was Moshe Ben-David honored for his contributions to Islamic literature?”), computed with R-TOFU’s reference all-MiniLM-L6-v2 encoder. Top: each method’s output— REDaCT declines in the answer after a two-sentence CoT, while R 2 MU collapses and emits no answer. Table: CFE scores the CoT —each row is a ground-truth CoT sentence and the bar is its best-matching generated CoT sentence’s cosine under REDaCT and under R 2 MU. REDaCT ’s CoT reuses the two opener sentences verbatim (cosine

1.0

) and stays topically close elsewhere through shared question words ( award , Islamic , literature ) without reproducing any biographical fact—it never emits “Distinguished”, “Tel Aviv”, or “Nobel”—so its dissimilarity is low, whereas R 2 MU’s punctuation collapse matches nothing and scores near-maximal dissimilarity. Because CFE credits dissimilarity, the collapse outscores REDaCT (

0.876

vs.

0.346

cosine;

0.989

vs.

0.650

ROUGE-L) even though REDaCT is the method that actually refuses—the inversion holds on

96.2\%

of forget10 prompts.

Method	Learning rate	Epochs / steps
GA	$6\times 10^{-6}$	2 epochs
GD	$6\times 10^{-6}$	3 epochs
KL	$6\times 10^{-6}$	3 epochs
PO	$4\times 10^{-6}$	2 epochs

Appendix

Table 4: Baseline hyperparameters selected on the DeepSeek-R1-Distill-Qwen-7B target for GA, GD, KL, PO.

Method	AFE $\uparrow$	CFE $\uparrow$	Utility $\uparrow$	Halluc. $\downarrow$	Refusal $\uparrow$	Collapse $\downarrow$
Target (pre-unlearning)	—	—	0.652	—	—	—
GA	0.453	0.689	0.671	0.950	0.044	0.006
GD	0.508	0.630	0.565	0.985	0.010	0.005
KL	0.805	0.965	0.592	0.419	0.115	0.466
PO	0.350	0.409	0.545	0.571	0.400	0.029
R 2 MU	0.990	0.992	0.381	0.066	0.028	0.907

Appendix

Table 5: Performance of REDaCT and baseline approaches on the DeepSeek-R1-Distill-Qwen-7B target ( forget10 ). Columns as in Table 1 ; bold values mark the best value per column.

Forget Set	Recall	Precision	F1
forget01	1.000	1.000	1.000
forget05	1.000	0.995	0.998
forget10	0.995	1.000	0.998

Appendix

Table 6: CoT-level recall, precision, and F1 of the leak detectors at

\tau=0.5

Panel of Figure 7 — Figure 7: Impact of Corrective embedding length on performance of REDaCT

Machine Unlearning

Machine unlearning removes the influence of selected data or concepts from trained models while preserving useful behavior and avoiding full retraining.

See all 349 papers

GUARD: Natural Forgetting in Large Reasoning Models via Guided Answer-Reasoning Distillation
Sep 18, 2026Zeyu Yan, Guanghao Zhou, Minghui Qiu +2Privacy Leakage in Language Models Machine Unlearning
East China Normal University, Shanghai, China · Unaffiliated
CiPO: Counterfactual Unlearning for Large Reasoning Models through Iterative Preference Optimization
Apr 17, 2026Junyi Li, Yongqiang Chen, Ningning DingMachine Unlearning LLM Unlearning
The Hong Kong University of Science and Technology (Guangzhou) · The Chinese University of Hong Kong
Selective Forgetting for Large Reasoning Models
Apr 4, 2026Tuan Le, Wei Qian, Mengdi HuaiLLM Unlearning Large Reasoning Models
Iowa State University, Ames IA 50011, USA

Sep 18, 2026J/K move · Enter open · S save

GUARD: Natural Forgetting in Large Reasoning Models via Guided Answer-Reasoning Distillation

Zeyu Yan, Guanghao Zhou, Minghui Qiu, Ming Gao+1

East China Normal University, Shanghai, China · Unaffiliated

Privacy Leakage in Language Models Machine Unlearning LLM Unlearning Large Reasoning Models

Open paper AskPDF arXiv

Recent advances in large reasoning models (LRMs) have made machine unlearning more challenging, as protected facts or unsafe rationales may surface in intermediate chain-of-thought (CoT) traces before the final answer is produced. Existing unlearning objectives typically suppress the target content or redirect internal representations, but they never specify how the post-forgetting trajectory should continue, which can lead to hallucinated substitutes, malformed boundaries, or repetitive outputs. We argue that LRM unlearning should instead learn a natural forgetting trajectory: a coherent non-disclosing CoT followed by a stable refusal-style answer that replace the original disclosure. To this end, we propose Guided Answer-Reasoning Distillation (GUARD), which converts model-generated unsafe disclosures into safe-exit trajectories, aligns a frozen LRM via guidance tokens, and distills the guided behavior into model parameters. To address the lack of metrics for replacement quality beyond leakage, we further introduce Natural Forgetting Reasoning Score (NFRS), which captures structural stability, fluency, and unsupported substitutes in forgotten outputs. Extensive experiments on R-TOFU and a STAR-1-derived harmful-intent setting show that GUARD substantially reduces unsafe and privacy disclosures across two widely adopted distilled LRMs while preserving reasoning utility.

Abstract

Machine Unlearning

Explore similar work

GUARD: Natural Forgetting in Large Reasoning Models via Guided Answer-Reasoning Distillation

CiPO: Counterfactual Unlearning for Large Reasoning Models through Iterative Preference Optimization

Selective Forgetting for Large Reasoning Models