惡意微調防線被持續訓練擊穿

Computer Science > Cryptography and Security arXiv:2605.

14605 (cs) [Submitted on 14 May 2026 (v1), last revised 7 Oct 2026 (this version, v3)] Title:A Few Steps Further: Why Defenses Against Malicious Finetuning Erode Under Continued Training Authors:Itay Zloczower, Eyal Lenga, Gilad Gressel, Yisroel Mirsky View a PDF of the paper titled A Few Steps Further: Why Defenses Against Malicious Finetuning Erode Under Continued Training, by Itay Zloczower and 3 other authors View PDF HTML (experimental) Abstract:Model providers increasingly release the weights of large language models.

Although these models are safety-aligned before release, their safeguards can often be removed by fine-tuning on harmful data.

A growing class of defenses aims to make alignment robust to such malicious fine-tuning, but these defenses are typically evaluated against attacks with a fixed training budget, even though an attacker who holds the weights can simply train for longer.

We ask whether current defenses withstand this simplest escalation.

Surveying fifteen recent defenses, we find that they share a common weakness: each is built around a limited model of the attacker, such as a bounded perturbation, a short simulated attack, or a trained link between harmful and benign behavior, and nothing enforces that protection once the weights are released.

We then test six representative defenses on four open-weight models by continuing the same harmful-only fine-tuning attack for three epochs and measuring harmfulness and capability along the way.

In all 72 defended runs, the model is more harmful at the end of training than at release, and on Llama-3.1 at the highest learning rate the defended models end almost as harmful as the undefended one.

The defenses are not equally weak: one defense kept harmfulness low on one model, and some attacks recovered harmfulness only at the cost of general capability.

Current defenses can delay or disrupt malicious fine-tuning, but in most cases their measured resistance does not persist under continued training, and they should not yet be treated as durable protection.Subjects: Cryptography and Security (cs.CR); Artificial Intelligence (cs.

AI); Machine Learning (cs.LG) Cite as: arXiv:2605.14605 [cs.CR] (or arXiv:2605.14605v3 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2605.

14605 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Itay Zloczower [view email] [v1] Thu, 14 May 2026 09:22:14 UTC (199 KB) [v2] Sun, 24 May 2026 08:34:13 UTC (199 KB) [v3] Wed, 7 Oct 2026 16:55:56 UTC (265 KB) Full-text links: Access Paper: View a PDF of the paper titled A Few Steps Further: Why Defenses Against Malicious Finetuning Erode Under Continued Training, by Itay Zloczower and 3 other authorsView PDFHTML (experimental)TeX Source view license Additional Features Audio Summary Current browse context: cs.

CR < prev | next > new | recent | 2026-05 Change to browse by: cs cs.AI cs.LG References & Citations NASA ADSGoogle Scholar Semantic Scholar export BibTeX citation Loading...BibTeX formatted citation × loading...

Data provided by: Bookmark Bibliographic Tools Bibliographic and Citation Tools Bibliographic Explorer Toggle Bibliographic Explorer (What is the Explorer?) Connected Papers Toggle Connected Papers (What is Connected Papers?) Litmaps Toggle Litmaps (What is Litmaps?) scite.

ai Toggle scite Smart Citations (What are Smart Citations?) Code, Data, Media Code, Data and Media Associated with this Article alphaXiv Toggle alphaXiv (What is alphaXiv?) Links to Code Toggle CatalyzeX Code Finder for Papers (What is CatalyzeX?) DagsHub Toggle DagsHub (What is DagsHub?

) GotitPub Toggle Gotit.pub (What is GotitPub?) Huggingface Toggle Hugging Face (What is Huggingface?) ScienceCast Toggle ScienceCast (What is ScienceCast?) Demos Demos Replicate Toggle Replicate (What is Replicate?) Spaces Toggle Hugging Face Spaces (What is Spaces?) Spaces Toggle TXYZ.

AI (What is TXYZ.AI?) Related Papers Recommenders and Search Tools Link to Influence Flower Influence Flower (What are Influence Flowers?) Core recommender toggle CORE Recommender (What is CORE?

) Author Venue Institution Topic About arXivLabs arXivLabs: experimental projects with community collaborators arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy.arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community?Learn more about arXivLabs.Which authors of this paper are endorsers?| Disable MathJax (What is MathJax?)