[2506.17209] Fine-Tuning Lowers Safety and Disrupts Evaluation Consistency

Abstract:Fine-tuning a general-purpose large language model (LLM) for a specific domain or task has become a routine procedure for ordinary users. However, fine-tuning is known to remove the safety alignment features of the model, even when the fine-tuning data does not contain any harmful content. We consider this to be a critical failure mode of LLMs due to the widespread uptake of fine-tuning, combined with the benign nature of the "attack". Most well-intentioned developers are likely unaware that they are deploying an LLM with reduced safety. On the other hand, this known vulnerability can be easily exploited by malicious actors intending to bypass safety guardrails. To make any meaningful progress in mitigating this issue, we first need reliable and reproducible safety evaluations. In this work, we investigate how robust a safety benchmark is to trivial variations in the experimental procedure, and the stochastic nature of LLMs. Our initial experiments expose surprising variance in the results of the safety evaluation, even when seemingly inconsequential changes are made to the fine-tuning setup. Our observations have serious implications for how researchers in this field should report results to enable meaningful comparisons in the future.

# link_1c2xcoywhvl.pdf ## Metadata - PDFFormatVersion=1.5 - IsLinearized=false - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - Author=Kathleen C. Fraser; Hillary Dawkins; Isar Nejadgholi; Svetlana Kiritchenko - Creator=arXiv GenPDF (tex2pdf:) - Custom.DOI=https://doi.org/10.48550/arXiv.2506.17209 - Custom.License=http://creativecommons.org/licenses/by/4.0/ - Custom.PTEX.Fullbanner=This is pdfTeX, Version 3.141592653-2.6-1.40.25 (TeX Live 2023) kpathsea version 6.3.5 - Custom.arXivID=https://arxiv.org/abs/2506.17209v1 - Producer=pikepdf 8

saved by

• Evie Hu

related reading

• [2502.17424] Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org
• [2512.11949] Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitorsarxiv.org
• Fine-Tuning Lowers Safety and Disrupts Evaluation Consistencyarxiv.org
• Modifying LLM Beliefs with Synthetic Document Finetuningalignment.anthropic.com
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• [2502.05209] Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilitiesarxiv.org
• [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
• [2602.15799] The Geometry of Alignment Collapse: When Fine-Tuning Breaks Safetyarxiv.org
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• [2412.00543] Evaluating the Consistency of LLM Evaluatorsarxiv.org
• Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs 49 This paper contains model-generated content that might be offensive. 49arxiv.org
• Large Language Models Must Be Taught to Know What They Don't Knowarxiv.org