11-19 更新 实时
首页/ 算力/ 正文

三项新技术瞄准隐藏的AI后门且不损性能

算力 来源:TechXplore 2026-11-07 03:00 0 阅读 机器翻译
三项新技术瞄准隐藏的AI后门且不损性能

由 Gaby Clark 编辑,Robert Egan 审校

本文已按照 Science X 的编辑流程与政策进行审核。编辑在确保内容可信度的同时,重点强调了以下属性:

添加到首选来源

研究人员展示了三种新技术,用于防范所谓的「后门攻击」——这类攻击会诱导 AI 模型产生恶意或不良行为。研究人员还提出了一个构想,可将后门攻击技术重新加以利用,为用户增加一层额外的安全保障。

后门攻击指的是有人秘密在 AI 模型中植入「触发器」,一旦该触发器被激活,就会迫使模型执行特定操作。

「关键在于,触发器不存在时,模型可以表现正常,这使得后门在普通使用或测试过程中难以被察觉,」该研究相关四篇论文的通讯作者、北卡罗来纳州立大学计算机科学助理教授 Jung-Eun Kim 表示。「但一旦触发器被激活,它就能让模型做出攻击者想要的行为,例如生成不良文本、执行恶意命令。」

「鉴于 AI 模型已被用于从金融交易到软件开发的方方面面,能否控制 AI 的行为就构成了一个重大的脆弱点,」Kim 说。

随着 AI 模型日益普及、被用于越来越多的应用场景,人们越来越需要「微调」模型,使其高效完成这些新任务。微调指的是对 AI 模型进行额外训练,其中会使用额外的数据集。这为后门攻击创造了新的可乘之机——攻击可以针对模型正在被微调以执行的特定功能进行定制。

“这是一个众所周知的漏洞,但后门攻击很难防范,因为在触发器被激活之前,它们很难被识别,”北卡罗来纳州立大学博士生、全部四篇论文的第一作者 Jianwei Li 表示。“我们这项工作的目标是在部署之前就防范后门攻击,从而让 AI 更安全、更值得信赖。”

为了解决后门方面的隐患,研究人员从三个不同层面考察了 AI 模型:特征、架构和权重。

“权重是构成模型的参数,”Li 说。“特征是模型内部对数据的表示。而架构则指模型内部的运行组件。

“这些都是 AI 模型不可或缺的要素,我们开发的技术对这三个层面都进行了修改,以消除或缓解后门攻击,同时不会显著牺牲性能,”Li 说。

研究人员还提出使用“秘密对齐”(secret alignment),即利用与后门攻击中所用触发器类似的触发器来执行特定的有益任务。例如,可以植入一个只有管理层用户才知道的触发器,赋予他们访问安全数据或控制功能的受限权限。

“总体而言,我们提出了一套全面的技术,可降低后门攻击的威胁,而且这些技术都可以立即实施,”Kim 说。“我们还提出了一种方法,将后门触发器重新利用为有用的功能。”

聚焦 AI 模型特性的论文《Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection》将在第 40 届神经信息处理系统年会(NeurIPS)上展示,该会议将于 12 月在澳大利亚悉尼、美国亚特兰大和法国巴黎举行。聚焦 AI 模型架构的论文《Backdoor Containment via Expert Quarantine and Shutdown in LLMs》也将在 NeurIPS 上展示,其合著者包括北卡罗来纳州立大学(NC State)博士生 Min-Seon Kim。

聚焦 AI 模型权重的论文《Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference》已在第 14 届国际学习表征会议(ICLR)上展示,该会议于 4 月 23 日至 27 日在巴西里约热内卢举行。提出“秘密对齐”(secret alignment)概念的论文《Position: Retire the 'Positive Backdoor' Label—Secret Alignment Requires Strict and Systematic Evaluation》已在第 43 届国际机器学习大会(ICML)上展示,该会议于 7 月 6 日至 11 日在韩国首尔举行。该研究已在 arXiv 上发布预印本。

Jianwei Li et al., Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection, arXiv (2026). DOI: 10.48550/arxiv.2610.00685

Jianwei Li et al., Backdoor Containment via Expert Quarantine and Shutdown in LLMs, arXiv (2026). DOI: 10.48550/arxiv.2610.00663

Jianwei Li et al., Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference, arXiv (2026). DOI: 10.48550/arxiv.2603.13461

Jianwei Li et al., Position: Retire the "Positive Backdoor" Label—Secret Alignment Requires Strict and Systematic Evaluation, arXiv (2026). DOI: 10.48550/arxiv.2605.28597

期刊信息:arXiv

这则报道的背后是谁?

英语硕士,自 2021 年起担任文字编辑,具备高等教育与健康领域内容经验。致力于可信赖的科学新闻。完整简介 →

数学生物学学士,创意写作硕士。游历广泛,对科学与语言有独到见解。完整简介 →

引用来源:Three new techniques target hidden AI backdoors without major performance losses(2026 年 11 月 6 日),检索于 2026 年 11 月 6 日,来自 https://techxplore.com/news/2026-11-techniques-hidden-ai-backdoors-major.html

本文件受版权保护。除为私人学习或研究目的进行的任何合理使用外,未经书面许可,不得复制本文件的任何部分。内容仅供参考之用。

获取即时摘要文本(要点)

三种分别针对 AI 模型特征、架构和权重的技术移除或缓解了隐藏后门,且未造成实质性性能损失。另一项独立提案将利用秘密触发器来限制对敏感数据或控制功能的访问。

本摘要由 LLM 自动生成。完整免责声明

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Three new techniques target hidden AI backdoors without major performance losses)

edited by Gaby Clark, reviewed by Robert Egan

This article has been reviewed according to Science X's editorial process and policies. Editors have highlighted the following attributes while ensuring the content's credibility:

Add to Preferred Sources

A man looks at two computer screens covered in code Credit: Jefferson Santos

Researchers have demonstrated three new techniques to prevent so-called "backdoor attacks," which induce malicious or undesirable behavior in AI models. The researchers have also proposed a concept that would repurpose backdoor attack techniques to create an additional layer of security for users.

Backdoor attacks refer to instances in which someone has secretly embedded a "trigger" in an AI model that—when activated—forces the model to perform specific actions.

"The important point is that the model can behave normally when the trigger is absent, making the backdoor difficult to notice during ordinary use or testing," says Jung-Eun Kim, corresponding author of four papers describing the work and an assistant professor of computer science at North Carolina State University. "But when the trigger is activated, it can cause the model to do things an attacker intended, such as generating unwanted text and executing malicious commands.

"Given that AI models are used for everything from financial transactions to software development, the ability to control behaviors of AI represents a significant vulnerability," says Kim.

As AI models have become more common and are being used for more applications, there is an increasing need to "fine-tune" the models to perform these new tasks effectively. Fine-tuning refers to additional training for AI models, which involves using additional data sets. This creates new opportunities for backdoor attacks, which can be tailored to the specific functions that the model is being fine-tuned to perform.

"This is a well-known vulnerability, but backdoor attacks are difficult to prevent because they are hard to identify until the trigger has been activated," says Jianwei Li, first author of all four papers and a Ph.D. student at NC State. "Our goal with this work was to prevent backdoor attacks before deployment in order to make AI safer and more trustworthy."

To address backdoor concerns, the researchers looked at three different levels of AI models: features, architecture and weights.

"Weights are the parameters that make up a model," says Li. "Features are internal representations of the data in a model. And architecture refers to the internal operating components of the model.

"These are all essential elements of an AI model, and we've developed techniques that modify all three of those levels to remove or mitigate backdoor attacks without making significant sacrifices in performance," says Li.

The researchers also propose the use of "secret alignment," using triggers like those used in backdoor attacks to perform specific beneficial tasks. For example, a trigger could be installed that would be known only to management-level users, giving them restricted access to secure data or control functions.

"Overall, we've introduced a comprehensive suite of techniques that reduces the threat of backdoor attacks, all of which could be implemented immediately," says Kim. "And we've also outlined an approach for repurposing backdoor triggers into useful features."

The paper focusing on AI model features, "Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection," will be presented at the Fortieth Annual Conference on Neural Information Processing Systems (NeurIPS), being held in December in Sydney (Australia), Atlanta (U.S.) and Paris (France). The paper focusing on AI model architecture, "Backdoor Containment via Expert Quarantine and Shutdown in LLMs," will also be presented at NeurIPS and was co-authored by NC State Ph.D. student Min-Seon Kim.

The paper focusing on AI model weights, "Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference," was presented at the Fourteenth International Conference on Learning Representations (ICLR), held April 23–27 in Rio de Janeiro, Brazil. The paper proposing the concept of "secret alignment," "Position: Retire the 'Positive Backdoor' Label—Secret Alignment Requires Strict and Systematic Evaluation," was presented at the Forty-Third International Conference on Machine Learning (ICML), held July 6–11 in Seoul, South Korea. The research is available for preprint on arXiv.

Jianwei Li et al., Backdoor Purification for LoRA-Tuned LLMs via Null-Space Projection, arXiv (2026). DOI: 10.48550/arxiv.2610.00685

Jianwei Li et al., Backdoor Containment via Expert Quarantine and Shutdown in LLMs, arXiv (2026). DOI: 10.48550/arxiv.2610.00663

Jianwei Li et al., Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference, arXiv (2026). DOI: 10.48550/arxiv.2603.13461

Jianwei Li et al., Position: Retire the "Positive Backdoor" Label—Secret Alignment Requires Strict and Systematic Evaluation, arXiv (2026). DOI: 10.48550/arxiv.2605.28597

Journal information:arXiv

Who's behind this story?

Gaby Clark

MA in English, copy editor since 2021 with experience in higher education and health content. Dedicated to trustworthy science news. Full profile →

Bachelor's in mathematical biology, Master's in creative writing. Well-traveled with unique perspectives on science and language. Full profile →

Citation: Three new techniques target hidden AI backdoors without major performance losses (2026, November 6) retrieved 6 November 2026 from https://techxplore.com/news/2026-11-techniques-hidden-ai-backdoors-major.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.

Get Instant Summarized Text (Gist)

Three techniques targeting AI model features, architecture and weights removed or mitigated hidden backdoors without substantial performance losses. A separate proposal would use secret triggers to restrict access to sensitive data or control functions.

This summary was automatically generated using LLM. Full disclaimer

This site uses cookies to assist with navigation, analyse your use of our services, collect data for ads personalisation and provide content from third parties. By using our site, you acknowledge that you have read and understand our Privacy Policy and Terms of Use.

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 TechXplore(Three new techniques target hidden AI backdoors without major performance losses),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗