- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes PACE, a persona unlearning method that aims to remove undesirable behavioral patterns from LLMs while preserving their general capabilities and response quality. The main idea is to identify persona-specific directions in the hidden representation space by contrasting target and desirable responses to the same questions. PACE then fine-tunes the model to suppress the target persona direction while aligning the remaining representations with those of desirable counterparts. The paper also introduces PersonaUnlearnBench, covering six LLMs and five personas. Experiments show that PACE generally outperforms existing unlearning baselines in suppressing undesirable personas while maintaining response quality, although some degradation in general utility remains.
Critical strengths
- The paper introduces persona unlearning as a distinct task from conventional knowledge unlearning, focusing on removing undesirable behavioral patterns rather than specific factual content. This provides an interesting perspective on improving the safety and controllability of LLMs.
- PACE is relatively simple and intuitive, using persona directions to suppress undesirable behaviors through representation-level optimization. The counterpart anchoring mechanism also helps preserve desirable behaviors and response quality during unlearning.
- The paper evaluates six LLMs from three model families across five personas and compares PACE with several existing unlearning baselines. It also includes ablation studies and robustness evaluations under jailbreak attacks, multi-turn interactions, and cross-lingual settings, which help support the effectiveness of the method.
Critical weaknesses
- The methodological novelty is relatively limited. PACE builds on existing Persona Vector and Activation Steering techniques, using persona directions and representation-level optimization to suppress undesirable behaviors. Although counterpart anchoring is introduced, the overall approach is closely related to existing representation editing methods.
- Persona unlearning can still degrade general model capabilities. Although PACE effectively suppresses target personas, it causes noticeable drops in general utility in some settings. For example, on Llama-3.1-8B with the Impolite persona, GU decreases from 68.65 to 63.75, raising concerns about the trade-off between persona removal and capability preservation.
- The long-term robustness of persona unlearning is not sufficiently evaluated. The paper tests jailbreak attacks, multi-turn interactions, and cross-lingual settings, but does not examine whether forgotten personas can re-emerge after subsequent fine-tuning. Moreover, the forgetting performance decreases under iterative jailbreak attacks for some personas, raising concerns about the persistence of the unlearning effect.
Questions for authors
Have the authors considered evaluating whether the forgotten personas can re-emerge after subsequent fine-tuning? Since PACE aims to achieve persistent persona unlearning through weight-level modifications, such experiments would help assess the long-term effectiveness and robustness of the proposed method.
6 comments
Response to Reviewer hRxY (Part 2/3)
Weakness 2: General-utility loss on Llama-3.1-8B–Impolite
“Persona unlearning can still degrade general model capabilities. Although PACE effectively suppresses target personas, it causes noticeable drops in general utility in some settings. For example, on Llama-3.1-8B with the Impolite persona, GU decreases from 68.65 to 63.75, raising concerns about the trade-off between persona removal and capability preservation.”
Thank you for highlighting this trade-off. We clarify that: (1) on Llama–Impolite, the 4.90-point GU change accompanies complete target suppression (+96.30 TF) and a +19.10-point improvement in response quality, (2) core mathematical and scientific reasoning on GSM8K and ARC-Challenge remain well preserved, and (3) at nearly the exact same observed GU (63.75 vs. 63.81), PaCE outperforms direct task-matched SFT by +14.08 TF and +21.45 RQ.
-
Substantial behavioral and response-quality gains
Table C.2: Llama-3.1-8B–Impolite behavioral gains and utility cost.
Metric Meaning Original SFT (Matched) PaCE PaCE vs. Orig PaCE vs. SFT TF ↑ Target-persona suppression 3.70 85.92 100.00 +96.30 +14.08 CP ↑ Constructive-counterpart expression 74.60 75.60 73.60 −1.00 −2.00 RQ ↑ Quality of persona-task answers 74.10 71.75 93.20 +19.10 +21.45 GU ↑ Mean general-capability score 68.65 63.81 63.75 −4.90 −0.06 As shown in Table C.2, TF increases by 96.30 points and RQ increases by 19.10 points, while CP remains within 1.00 point of Original.
-
Breakdown across general-utility capabilities
Table C.3: General-utility breakdown on Llama-3.1-8B–Impolite.
Benchmark Capability Test items Original PaCE Change GSM8K Mathematical word-problem reasoning 1,319 84.15 83.62 −0.53 ARC-Challenge Science question answering and reasoning 1,172 53.75 51.11 −2.64 MMLU Broad multitask knowledge 14,042 63.12 57.22 −5.90 IFEval Following explicit formatting instructions 541 73.57 63.03 −10.54 GU Equal-weight mean — 68.65 63.75 −4.90 Table C.3 shows that mathematical reasoning on GSM8K is largely unchanged (84.15 → 83.62; −0.53 points) and ARC-Challenge changes by only −2.65 points, whereas the decrease is concentrated in strict formatting adherence on IFEval.
-
Comparison with direct SFT at nearly identical observed GU
When we train Llama-3.1-8B on the same 400 pairs and 75 updates using task-matched SFT (training to produce the constructive response under the target prompt), SFT yields virtually the same GU (63.81 vs. PaCE’s 63.75), yet PaCE achieves:
- TF 100.00 vs. 85.92 (+14.08 points), and
- RQ 93.20 vs. 71.75 (+21.45 points).
Moreover, across all 12 model–persona settings (Appx. F.8, Table 13), PaCE achieves higher TF than both SFT and DPO in 12/12 settings and higher RQ in 10/12 settings, and on Qwen3.5-9B it simultaneously surpasses both SFT and DPO on TF, RQ, and GU across all four personas.
Response to Reviewer hRxY (Part 3/3)
Weakness 3: Persistence after fine-tuning and residual accessibility under attacks
“The long-term robustness of persona unlearning is not sufficiently evaluated. The paper tests jailbreak attacks, multi-turn interactions, and cross-lingual settings, but does not examine whether forgotten personas can re-emerge after subsequent fine-tuning. Moreover, the forgetting performance decreases under iterative jailbreak attacks for some personas, raising concerns about the persistence of the unlearning effect.”
Thank you for distinguishing prompt-level robustness from persistence under subsequent training. We address this by adding a downstream full-parameter fine-tuning experiment on FinQA [1] without persona supervision (Table C.4). Three personas retain their exact starting TF across all measured checkpoints, and all four maintain a large endpoint advantage over the equally fine-tuned Original model.
-
Downstream full-parameter adaptation on FinQA without persona supervision
We fine-tune the Original and four PaCE-edited Llama-3.1-8B checkpoints for 128 full-parameter optimizer updates on 1,024 FinQA examples [1] (numerical reasoning over financial reports). Training uses standard supervised loss alone, with neither persona data nor the PaCE objective. We evaluate the same 50 target-persona questions per setting at steps 0, 32, and 128 (0.5-point TF resolution).
Table C.4: TF (↑) during downstream full-parameter FinQA fine-tuning.
Persona Original: 0 Original: 32 Original: 128 PaCE: 0 PaCE: 32 PaCE: 128 Endpoint Gain Sycophantic 0.0 1.5 1.0 94.0 94.0 94.0 +93.0 Hallucinating 15.5 22.5 19.5 86.5 85.0 80.0 +60.5 Impolite 2.0 6.0 4.0 100.0 100.0 100.0 +96.0 Apathetic 66.5 76.0 69.0 100.0 100.0 100.0 +31.0 -
Suppression persists across downstream weight updates
- Sycophantic remains at 94.0, while Impolite and Apathetic remain at 100.0 across all three checkpoints (steps 0, 32, and 128).
- Hallucinating retains TF 80.0 at step 128, remaining +60.5 points above the equally fine-tuned Original model (19.5).
- Combined with our multi-turn (86.5–100.0 TF across 5 turns), cross-lingual (100.0 TF across 5 languages), and iterative Attack@5 evaluations (where Impolite, Apathetic, and Evil retain 100.0, 99.0, and 91.5 TF, and Sycophantic/Hallucinating still substantially outperform all baselines), these results show that PaCE’s weight edit remains durable well beyond the immediate post-unlearning checkpoint. We have added this study to the revised manuscript (Sec. 5.3, Table 2 and Appx. F.9).
[1] Chen et al. “FinQA: A Dataset of Numerical Reasoning over Financial Data”. In EMNLP, 2021.
Question: Can the forgotten personas re-emerge after subsequent fine-tuning?
“Have the authors considered evaluating whether the forgotten personas can re-emerge after subsequent fine-tuning? Since PACE aims to achieve persistent persona unlearning through weight-level modifications, such experiments would help assess the long-term effectiveness and robustness of the proposed method.”
Yes, we have added a direct test of subsequent full-parameter fine-tuning (Table C.4 in Weakness 3, included in Sec. 5.3, Table 2 and Appx. F.9 of the revised paper). After 128 full-parameter updates on 1,024 FinQA examples without persona supervision or the PaCE loss, three edited personas (Sycophantic at 94.0, Impolite at 100.0, and Apathetic at 100.0) retain their exact starting TF at all checkpoints, and Hallucinating retains 80.0 TF (+60.5 points above the fine-tuned Original’s 19.5), demonstrating strong persistence under downstream weight adaptation.
Response to Reviewer hRxY (Part 1/3)
Thank you for your positive feedback on the relevance of persona unlearning to LLM safety and controllability, the simplicity of PaCE, and the breadth of our experiments. Below we address each of your weaknesses and questions point-by-point.
Weakness 1: Relationship to existing representation-editing methods
Thank you for raising this point. We clarify that: (1) unlike prior LLM unlearning methods that focus on memorized content, our work is the first to formulate and study weight-level behavioral unlearning, combining the linear representation idea of persona vectors with a simple, direct, and hook-free parameter-editing objective; and (2) our controlled component study (Table C.1) verifies that PaCE’s three specific design choices—pre-generation prompt-state readout, per-question counterpart threshold, and orthogonal anchoring—each provide clear empirical gains over simpler alternatives.
Distinction from content unlearning and inference-time steering
Existing LLM unlearning methods remove specific memorized facts or texts, which often fails to suppress a recurring behavioral persona without degrading generation quality (Sec. 3.5). Meanwhile, Activation Steering and Persona Vectors [1] primarily study inference-time hooks (which can be removed in open-weight settings) or response-token regularization.
To achieve persistent behavioral unlearning in model weights without runtime hooks, PaCE uses the frozen Original model’s paired counterpart on the same question to define a two-part representation edit at the final prompt token before generation:
L=Ei[sit−sic+m]++λEi[1−cos(P⊥hit,P⊥hi,0c)],[z]+=max(z,0).
Here, the erasure term pushes the target projection sit=v⊤hit below the question-specific counterpart projection sic=v⊤hi,0c by margin m, while the anchor aligns the remaining representation strictly in the orthogonal subspace P⊥=I−vv⊤ so preservation does not conflict with erasure.
Controlled evaluation of readout position, threshold calibration, and orthogonal anchoring
We compare five variants (
A0–A4) on Gemma-4-12B / Sycophantic under the same 400 paired examples, 75 optimizer updates, and fixed parameters (m=1,λ=1):A2) and the full-state anchor (A3).Table C.1: Controlled alternatives to PaCE’s design (Gemma-4-12B, Sycophantic).
Table C.1 shows that A0 achieves the highest TF and RQ among all five variants:
A0vs.A1) improves all four metrics: +3.50 TF, +0.50 CP, +14.40 RQ, and +0.77 GU.A0vs.A2) gains +4.90 TF and +6.00 RQ over a global threshold.A0vs.A3) yields the largest TF gain (+11.00 TF and +5.10 RQ over full-state anchoring, and +8.20 TF / +6.40 RQ overA4).We have included these results and clarifications in the revised manuscript (Sec. 1, Sec. 5.4 Table 3, and Appx. B.3).