Rethinking Safety in LLM Fine-tuning: An Optimization Perspective

• URL: http://arxiv.org/abs/2508.12531v1
• Date: Sun, 17 Aug 2025 23:46:36 GMT
• Title: Rethinking Safety in LLM Fine-tuning: An Optimization Perspective
• Authors: Minseon Kim, Jin Myung Kwak, Lama Alssum, Bernard Ghanem, Philip Torr, David Krueger, Fazl Barez, Adel Bibi,
• Abstract summary: We show that poor optimization choices, rather than inherent trade-offs, often cause safety problems, measured as harmful responses to adversarial prompts.<n>We propose a simple exponential moving average (EMA) momentum technique in parameter space that preserves safety performance.<n>Our experiments on the Llama families across multiple datasets demonstrate that safety problems can largely be avoided without specialized interventions.
• Score: 56.31306558218838
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: Fine-tuning language models is commonly believed to inevitably harm their safety, i.e., refusing to respond to harmful user requests, even when using harmless datasets, thus requiring additional safety measures. We challenge this belief through systematic testing, showing that poor optimization choices, rather than inherent trade-offs, often cause safety problems, measured as harmful responses to adversarial prompts. By properly selecting key training hyper-parameters, e.g., learning rate, batch size, and gradient steps, we reduce unsafe model responses from 16\% to approximately 5\%, as measured by keyword matching, while maintaining utility performance. Based on this observation, we propose a simple exponential moving average (EMA) momentum technique in parameter space that preserves safety performance by creating a stable optimization path and retains the original pre-trained model's safety properties. Our experiments on the Llama families across multiple datasets (Dolly, Alpaca, ORCA) demonstrate that safety problems during fine-tuning can largely be avoided without specialized interventions, outperforming existing approaches that require additional safety data while offering practical guidelines for maintaining both model performance and safety during adaptation.

Related papers

• Gradient Surgery for Safe LLM Fine-Tuning [16.652518818576425]
  Fine-tuning-as-a-Service introduces a critical vulnerability where a few malicious examples mixed into the user's fine-tuning dataset can compromise the safety alignment of Large Language Models (LLMs)<n>We find existing solutions are critically sensitive to the harmful ratio, with defenses degrading sharply as harmful ratio increases.<n>We propose SafeGrad, a novel method that employs gradient surgery.
  arXiv  Detail & Related papers  (2025-08-10T04:13:41Z)
• Shape it Up! Restoring LLM Safety during Finetuning [66.46166656543761]
  Finetuning large language models (LLMs) enables user-specific customization but introduces critical safety risks.<n>We propose dynamic safety shaping (DSS), a framework that uses fine-grained safety signals to reinforce learning from safe segments of a response while suppressing unsafe content.<n>We present STAR-DSS, guided by STAR scores, that robustly mitigates finetuning risks and delivers substantial safety improvements across diverse threats, datasets, and model families.
  arXiv  Detail & Related papers  (2025-05-22T18:05:16Z)
• Mitigating Fine-tuning Risks in LLMs via Safety-Aware Probing Optimization [7.1060720569792215]
  Fine-tuning large language models (LLMs) can inadvertently compromise their safety.<n>We introduce a safety-aware probing (SAP) framework designed to mitigate the safety risks.<n>Our experimental results demonstrate that SAP effectively reduces harmfulness below the original fine-tuned model.
  arXiv  Detail & Related papers  (2025-05-22T14:52:10Z)
• LookAhead Tuning: Safer Language Models via Partial Answer Previews [62.529794567687354]
  Fine-tuning enables large language models to adapt to specific domains, but often compromises their previously established safety alignment.<n>We introduce LookAhead Tuning, a lightweight and effective data-driven approach that preserves safety during fine-tuning.
  arXiv  Detail & Related papers  (2025-03-24T18:11:42Z)
• Vulnerability Mitigation for Safety-Aligned Language Models via Debiasing [12.986006070964772]
  Safety alignment is an essential research topic for real-world AI applications.<n>Our study first identified the difficulty of eliminating such vulnerabilities without sacrificing the model's helpfulness.<n>Our method could enhance the model's helpfulness while maintaining safety, thus improving the trade-off-front.
  arXiv  Detail & Related papers  (2025-02-04T09:31:54Z)
• Safeguard Fine-Tuned LLMs Through Pre- and Post-Tuning Model Merging [47.33307521558814]
  Fine-tuning large language models (LLMs) for downstream tasks often leads to catastrophic forgetting.<n>We show that simply merging the weights of pre- and post-fine-tuned models effectively mitigates safety degradation while enhancing performance.
  arXiv  Detail & Related papers  (2024-12-27T08:03:22Z)
• Safe to Serve: Aligning Instruction-Tuned Models for Safety and Helpfulness [0.0]
  Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning and text generation.<n>LLMs can inadvertently generate unsafe or biased responses when prompted with problematic inputs.<n>This research addresses the critical challenge of developing language models that generate both helpful and harmless content.
  arXiv  Detail & Related papers  (2024-11-26T06:52:22Z)
• What Makes and Breaks Safety Fine-tuning? A Mechanistic Study [64.9691741899956]
  Safety fine-tuning helps align Large Language Models (LLMs) with human preferences for their safe deployment. We design a synthetic data generation framework that captures salient aspects of an unsafe input. Using this, we investigate three well-known safety fine-tuning methods.
  arXiv  Detail & Related papers  (2024-07-14T16:12:57Z)
• Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training [67.30423823744506]
  We introduce a novel approach, Decoupled Refusal Training (DeRTa), designed to empower LLMs to refuse compliance to harmful prompts at any response position.<n>DeRTa incorporates two novel components: (1) Maximum Likelihood Estimation with Harmful Response Prefix, which trains models to recognize and avoid unsafe content by appending a segment of harmful response to the beginning of a safe response, and (2) Reinforced Transition Optimization (RTO), which equips models with the ability to transition from potential harm to safety refusal consistently throughout the harmful response sequence.
  arXiv  Detail & Related papers  (2024-07-12T09:36:33Z)
• Meta-Learning Priors for Safe Bayesian Optimization [72.8349503901712]
  We build on a meta-learning algorithm, F-PACOH, capable of providing reliable uncertainty quantification in settings of data scarcity. As core contribution, we develop a novel framework for choosing safety-compliant priors in a data-riven manner. On benchmark functions and a high-precision motion system, we demonstrate that our meta-learned priors accelerate the convergence of safe BO approaches.
  arXiv  Detail & Related papers  (2022-10-03T08:38:38Z)
• Log Barriers for Safe Black-box Optimization with Application to Safe Reinforcement Learning [72.97229770329214]
  We introduce a general approach for seeking high dimensional non-linear optimization problems in which maintaining safety during learning is crucial. Our approach called LBSGD is based on applying a logarithmic barrier approximation with a carefully chosen step size. We demonstrate the effectiveness of our approach on minimizing violation in policy tasks in safe reinforcement learning.
  arXiv  Detail & Related papers  (2022-07-21T11:14:47Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.