11-19 更新 实时
首页/ 算力/ 正文

研究者呼吁从生态入手,终结AI深度伪造打地鼠战

算力 来源:TechXplore 2026-11-10 04:20 0 阅读 机器翻译
研究者呼吁从生态入手,终结AI深度伪造打地鼠战

由 Lisa Lock 编辑,Andrew Zinin 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,重点标注了以下属性:

添加到首选来源

监管 AIG-NCII 生态系统的美国联邦与州法律全景。图片来源:Proceedings of the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (2026)。DOI: 10.1145/3846167.3848604

2025 年 5 月,《TAKE IT DOWN 法案》成为首部针对深度伪造及其他形式 AI 生成非自愿私密影像的联邦法律。该法律将传播“复仇式色情”(包括 AI 生成的素材)定为犯罪,并要求社交媒体平台及其他网站一经请求即迅速删除相关素材。

布朗大学研究人员发表在 Proceedings of the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization 上的一项新分析发现,这部联邦深度伪造法律——以及此前和此后出台的各州法律——都未能充分应对支撑这类 AI 生成性侵害的庞大数字生态系统。研究人员表示,该研究试图完整勾勒这一生态系统的全貌,以期为政策制定者及其他相关方更好地解决这一问题提供指引。

该研究由布朗大学技术责任、重塑与再设计中心的计算机科学家牵头,于 11 月 7 日星期六在 2026 年 ACM Conference on Equity and Access in Algorithms, Mechanisms and Optimization 上发布。研究团队成员与两名布朗大学本科生还创建了一个交互式配套网站,希望借此提高人们对 AI 助长的侵害行为及其所造成伤害的认识。

“核心问题在于,深度伪造滥用不仅仅涉及某一张图片、某个模型、某个网站或某个应用,而是涉及整个生态系统,”布朗大学计算机科学博士生、该研究第一作者 Michelle Ding 表示。“为了保护网络上的弱势群体免受这类滥用可能带来的各种伤害和情感痛苦,我们必须首先清晰地看清这个生态系统。”

深度伪造滥用背后的技术

AI 生成的非自愿私密图像并不算特别新鲜。该研究指出,早期的一次大规模扩散发生在 2017 年的 Reddit 上,当时人们利用换脸技术制作名人的裸照。生成式 AI 模型及其他技术的出现让这一现象急剧放大。2025 年,社交媒体网站 X 上的用户利用 Grok AI 制作了数千张性化图像,其中大多是女性和女孩。如今,能在几秒内生成露骨图片和视频的 AI“一键脱衣”应用已屡见不鲜——而且常常能在主流智能手机应用商店中下载到。

为了描绘这一新兴的 AI 生成非自愿私密图像生态系统,研究人员参考了学术文献、法律与政策报告、新闻报道及其他来源。最终形成了一套涵盖 11 个技术类别的分类体系,横跨深度伪造领域中的五种不同角色。

这些角色包括图像生成(涉及 AI 模型本身、用户界面和训练数据集等技术);分发传播(社交媒体和私密通讯应用);发现与扩散(搜索引擎、应用商店以及社交媒体上的深度伪造社群);基础设施与支持(GitHub、Civitai 等开发者平台);以及变现(信用卡公司和支付处理机构)。

为了理解这些角色和类别如何相互衔接,研究人员考察了由 X 的 Grok 所支撑的深度伪造生态系统。研究显示,Grok 基于 X 的界面让用户能够调用 Grok AI 模型(创作),只需在 X 帖子中 @Grok 并附上「给她穿上比基尼」或「脱掉她的衣服」之类的短语,就能生成未经同意的性化图像。Grok 则忠实地作出响应,直接在 X 的评论串中发布图像(传播)。

与此同时,Telegram 消息应用上冒出了一个社群(扩散与发现),用户在其中摸索出如何通过越狱一个可在应用商店中付费购买(支付处理方)的 Grok 界面,制作出更加露骨的图像和视频。

法律针对的是个人,而非基础设施

研究人员随后分析了 64 部涉及 AI 生成的非自愿图像的州级和联邦法律。他们发现,其中超过 92% 的法律针对的是滥用材料的个人制作者和传播者,而几乎未触及更大的技术生态系统。研究人员表示,这导致了一场无休止的「打地鼠」游戏,在网络上追逐个别的制作者和图像。

Ding 表示,这种零散的做法「根本不足以防止如此规模的滥用」。「去年 12 月,Grok 每小时生成超过 6,000 张图像。我们需要把关注点转向那些为这类滥用提供支持并从中获利的技术基础设施。」

研究人员希望他们的生态系统图谱能帮助立法者、监管机构和技术公司找到合适的施力点,从而更好地保护弱势群体。

“我们的目标之一,是推动研究去探索这些技术干预手段可能是什么样子,”布朗大学计算机科学助理教授、该研究合著者 Harini Suresh 说。“目前的缓解措施都非常被动——图像一旦流出就把它撤下来。在创作这一侧存在空白。比如,我们能否推动 AI 研究者去思考,如何从一开始就阻止模型生成这些图像?这是个难题,但也许我们能帮助把它提上研究议程。”

让生态系统变得可见

这项工作的配套网站旨在帮助把这一信息传递到科学界之外。研究团队与布朗大学两名本科生 Haya Bugshan 和 Maja Mishevska 合作,打造了一个互动体验,带领人们一次点击一步地了解深度伪造的整体图景。

“我们非常刻意地避免那种常常围绕深度伪造滥用报道的耸动渲染,”Mishevska 说。“我们选择让支撑这类滥用的技术和系统成为这一体验的焦点,通过互动叙事,读者可以自己去探索这些关联。”

其目的是帮助那些寻求解决方案的人在头脑中厘清关键概念。

“当我们谈论生态系统时,它会显得非常抽象,”Bugshan 说。“通过把这些技术呈现为人们在日常生活中会接触到的、可辨识的工具和网站,这种互动形式让生态系统变得更加具体可感,我认为这也让这项研究更易接近、更易理解。”

Michelle L. Ding 等,《如何停止打地鼠:绘制助长 AI 生成非自愿亲密图像的技术生态图谱》,第 6 届 ACM 算法、机制与优化中的公平与可及性会议论文集(2026)。DOI: 10.1145/3846167.3848604

由布朗大学提供

这篇报道的背后是谁?

艺术史学士,物质文化硕士。曾任博物馆编辑、急救员和器官移植协调员。自 2021 年起为 Science X 从事编辑工作。完整简介 →

物理学硕士,有科研经历。资深科学新闻爱好者。在 Science X 的编辑工作中发挥关键作用。完整简介 →

引用:为终结针对 AI 深度伪造的“打地鼠”式战争,研究人员呼吁从“生态”入手(2026 年 11 月 9 日),2026 年 11 月 9 日检索自 https://techxplore.com/news/2026-11-whack-mole-war-ai-deepfakes.html

本文件受版权保护。除出于私人学习或研究目的的合理使用外,未经书面许可不得复制任何部分。内容仅供信息参考之用。

获取即时摘要文本(要点)

一项分析梳理出 11 个技术类别,它们分属五个角色,共同支撑 AI 生成的未经同意的私密影像。在受审查的 64 部美国法律中,超过 92% 针对的是个人创作者和传播者,基本忽视了为其提供支持的基础设施。该分析建议采取更广泛的干预措施,针对纵容此类滥用行为并从中获利的整个生态体系。

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(To end 'whack-a-mole' war on AI deepfakes, address the 'ecosystem,' researchers urge)

edited by Lisa Lock, reviewed by Andrew Zinin

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

To end 'whack-a-mole' war on AI deepfakes, address the 'ecosystem," researchers urge The landscape of U.S. federal and state laws regulating the AIG-NCII ecosystem. Credit: Proceedings of the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (2026). DOI: 10.1145/3846167.3848604

In May 2025, the TAKE IT DOWN Act became the first federal law to address deepfakes and other forms of AI-generated nonconsensual intimate images. The law criminalizes the distribution of "revenge porn," including AI-generated material, and requires social media platforms and other websites to promptly remove the material upon request.

A new analysis by Brown University researchers published in the Proceedings of the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization finds that the federal deepfakes law—and the state laws that came before and after—do not adequately address the sprawling digital ecosystem that enables this form of AI-generated sexual abuse. The study attempts to fully map that ecosystem, the researchers say, in an effort to provide guidance for policymakers and others to better address the problem.

Led by computer scientists from Brown's Center for Technological Responsibility, Reimagination and Redesign, the study was presented on Saturday, Nov. 7, at the 2026 ACM Conference on Equity and Access in Algorithms, Mechanisms and Optimization. Members of the research team and a pair of Brown undergraduates also created an interactive companion website with the hope of raising awareness about AI-enabled abuse and the harms it causes.

"The core issue is that deepfake abuse is not just about one image, model, website or app, but an entire ecosystem," said Michelle Ding, a computer science Ph.D. student at Brown and the study's lead author. "In order to protect vulnerable communities online from the various types of harms and emotional distress that can come from this form of abuse, we must first see this ecosystem clearly."

The technologies behind deepfake abuse

AI-generated nonconsensual intimate images aren't especially new. The study notes that an early proliferation occurred in 2017 on Reddit, when people used face-swap technology to create nude images of celebrities. The emergence of generative AI models and other technologies has supercharged the phenomenon. In 2025, users on the social media site X used Grok AI to create thousands of sexual images of mostly women and girls. Today, AI "nudifier" apps, which can create explicit pictures and videos in seconds, are commonplace—often available in mainstream smartphone app stores.

To map this new and emerging ecosystem of AI-generated nonconsensual intimate images, the researchers drew upon the academic literature, law and policy reports, news stories and other sources. The result is a taxonomy of 11 technology categories that span five distinct roles in the deepfake landscape.

The roles include image creation (which involves technologies like the AI models themselves, user interfaces and training datasets); distribution (social media and private messaging apps); discovery and proliferation (search engines, app stores and deepfake communities on social media); infrastructure and support (developer platforms like GitHub and Civitai); and monetization (credit card companies and processors).

To understand how all of these roles and categories fit together, the researchers explored the deepfake ecosystem enabled by X's Grok. According to the research, Grok's X-based interface enabled users to access the Grok AI model (creation) to make nonconsensual sexual images just by tagging Grok in X posts with phrases like "put a bikini on her" or "remove her clothes." Grok dutifully responded by posting images directly in X comment threads (distribution).

Meanwhile, a community sprang up on the Telegram messaging app (proliferation and discovery) in which users figured out how to make even more explicit images and videos by jailbreaking a Grok interface available for purchase (payment processors) in app stores.

Laws focus on individuals, not infrastructure

The researchers went on to analyze 64 state and federal laws that address AI-generated nonconsensual images. They found that over 92% of them target individual creators and distributors of abuse material, while barely addressing the larger technological ecosystem. That, the researchers say, sets up an endless game of "whack-a-mole" that chases individual creators and images across the web.

That scattered approach is "simply not enough to prevent abuse at this scale," Ding said. "Grok was generating over 6,000 images every hour last December. We need to shift our focus to the technical infrastructure enabling and profiting from this abuse."

The researchers hope their ecosystem map will enable lawmakers, regulators and technology companies to find appropriate pressure points to better protect vulnerable populations.

"Part of our goal is to spur research into what these technical interventions might look like," said Harini Suresh, an assistant professor of computer science at Brown and study co-author. "Current mitigations are all very reactive—taking images down once they're out there. There's a gap on the creation side of things. Can we motivate AI researchers, for example, to think about how to prevent a model from creating these images in the first place? That's a tough problem, but maybe we can help to put that on the research agenda."

Making the ecosystem visible

The companion website for the work aims to help carry the message beyond the scientific community. The research team worked with two Brown undergraduates, Haya Bugshan and Maja Mishevska, to create an interactive experience that takes people through the deepfakes landscape click by click.

"We were very deliberate about avoiding the sensationalism that often surrounds coverage of deepfake abuse," Mishevska said. "We chose to make the technologies and systems enabling the abuse the focus of the experience, and through the interactive story, readers can explore those connections for themselves."

The goal is to help crystallize key concepts in the minds of those seeking solutions.

"When we talk about an ecosystem, it can feel very abstract," Bugshan said. "By presenting these technologies as recognizable tools and websites that people encounter in their everyday lives, the interactive format makes the ecosystem more tangible, which I think makes the research more accessible and easier to understand."

Michelle L. Ding et al., How to Stop Playing Whack-a-Mole: Mapping the Ecosystem of Technologies Facilitating AI-Generated Non-Consensual Intimate Images, Proceedings of the 6th ACM Conference on Equity and Access in Algorithms, Mechanisms, and Optimization (2026). DOI: 10.1145/3846167.3848604

Provided by Brown University

Who's behind this story?

Lisa Lock

BA art history, MA material culture. Former museum editor, paramedic, and transplant coordinator. Editing for Science X since 2021. Full profile →

Master's in physics with research experience. Long-time science news enthusiast. Plays key role in Science X's editorial success. Full profile →

Citation: To end 'whack-a-mole' war on AI deepfakes, address the 'ecosystem,' researchers urge (2026, November 9) retrieved 9 November 2026 from https://techxplore.com/news/2026-11-whack-mole-war-ai-deepfakes.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)

An analysis mapped 11 technology categories across five roles enabling AI-generated nonconsensual intimate imagery. Over 92% of 64 U.S. laws examined targeted individual creators and distributors, largely overlooking supporting infrastructure. The analysis recommends broader interventions targeting the ecosystem that enables and profits from abuse.

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(To end 'whack-a-mole' war on AI deepfakes, address the 'ecosystem,' researchers urge),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 ↗