SDD: Self-Degraded Defense against Malicious Fine-tuning

Zixuan Chen, Weikai Lu, Xin Lin, Ziqian Zeng
7/27/2025
cs.CRcs.AI

No official code found. Best implementation: ZeroNLP/SDD (1 star, Apache-2.0, last commit 2026-01-14).

Abstract

Open-source Large Language Models (LLMs) often employ safety alignment methods to resist harmful instructions. However, recent research shows that maliciously fine-tuning these LLMs on harmful data can easily bypass these safeguards. To counter this, we theoretically uncover why malicious fine-tuning succeeds and identify potential defense strategies. Building on the theoretical analysis, we introduce the Self-Degraded Defense (SDD) framework. SDD encourages LLMs to produce high-quality but irrelevant responses to harmful prompts. When attackers attempt malicious fine-tuning, the general capability of the LLM aligned by SDD will significantly decrease, rendering it incapable of following harmful instructions. Our experimental results confirm SDD's effectiveness against such attacks.

View on arXivView PDF

Code Implementations(1)

ZeroNLP/SDDCommunity68%
Apache-2.0

This is the official implementation of the paper "SDD: Self-Degraded Defense against Malicious Fine-tuning"

10PythonMay 30, 20259 months agoApache-2.0

Similar Papers

Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning

88% match1 repo102026

Targeted Vaccine: Safety Alignment for Large Language Models against Harmful Fine-Tuning via Layer-wise Perturbation

87% match1 repo192024

Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation

87% match1 repo142025

Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey

87% match1 repo2462024

No Two Devils Alike: Unveiling Distinct Mechanisms of Fine-tuning Attacks

87% match2024

Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence

87% match2026

Cite this paper

@article{chen2025sdd,
  title  = {SDD: Self-Degraded Defense against Malicious Fine-tuning},
  author = {Zixuan Chen and Weikai Lu and Xin Lin and Ziqian Zeng},
  year   = {2025},
  eprint = {2507.21182},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2507.21182v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2507.21182.svg)](https://codeofpaper.com/paper/2507.21182)

Discussion