Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability

• URL: http://arxiv.org/abs/2606.03648v1
• Date: Tue, 02 Jun 2026 13:39:17 GMT
• Title: Safety Measurements for Fine-tuned LLMs Should be Grounded in Capability
• Abstract summary: We conduct a multi-dimensional evaluation of the effects of fine-tuning on model behavior by focusing on capability as well as safety.<n>Our results surface important issues that (1) fine-tuned models can produce incoherent generations in response to safety prompts, (2) automated safety judgments are unreliable for such incoherent outputs, and (3) the conclusions about the effects of fine-tuning can change depending on the choice of safety benchmark as well as the safety evaluator.
• Score: 18.557410428487636
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Adapting foundation large language models to a user's task or preferred style through fine-tuning can result in compromising the model's safety. Previous works examined the effects of fine-tuning on model safety in limited and seemingly random experimental settings. We argue that anchoring fine-tuning to a specific capability goal is essential for avoiding arbitrary empirical choices, allowing us to draw meaningful conclusions about safety impacts, and to compare mitigation methods on a consistent basis. We conduct a multi-dimensional evaluation of the effects of fine-tuning on model behavior by focusing on capability as well as safety. Our results surface important issues that (1) fine-tuned models can produce incoherent generations in response to safety prompts, (2) automated safety judgments are unreliable for such incoherent outputs, and (3) the conclusions about the effects of fine-tuning can change depending on the choice of safety benchmark as well as the safety evaluator.

Related papers

• Rethinking Safety in LLM Fine-tuning: An Optimization Perspective [56.31306558218838]
  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.
  arXiv  Detail & Related papers  (2025-08-17T23:46:36Z)
• Shape it Up! Restoring LLM Safety during Finetuning [65.75757313781104]
  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)
• 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)
• 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)
• ProBF: Learning Probabilistic Safety Certificates with Barrier Functions [31.203344483485843]
  The control barrier function is a useful tool to guarantee safety if we have access to the ground-truth system dynamics. In practice, we have inaccurate knowledge of the system dynamics, which can lead to unsafe behaviors. We show the efficacy of this method through experiments on Segway and Quadrotor simulations.
  arXiv  Detail & Related papers  (2021-12-22T20:18:18Z)

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.