# 2026年9月AI训练数据版权诉讼判决及全球产业影响

> 针对生成式AI训练数据的版权争议，美欧正分别通过司法判例与立法监管，探索知识产权保护与技术创新的平衡秩序。

Published: 2026-09-19T06:34:25.283Z
Updated: 2026-09-19T06:34:25.283Z
URL: /zh/article/ai-copyright-training-2026

# 数据权利与算法边界：生成式AI版权争议与监管的分水岭

随着生成式人工智能（AI）技术跨越研究阶段、逐渐成为全产业链的核心基础设施，围绕用于训练大语言模型（LLM）的海量数据之法律属性的争议也达到了顶峰。将数据视作“新石油”并在网络上无差别抓取文本与图像的早期野蛮扩张时代已经落下帷幕。如今，司法判例与立法监管等精密的制度工具，正开始对AI训练数据的合法性与正当性展开实质性审查。

以美国为中心的“合理使用”（Fair Use）法理交锋，与欧盟（EU）成文化的监管体系，构成了全球数据治理的两大支柱。加之技术基础设施层面的控制手段以及商业数据许可模式的相继涌现，旨在实现知识产权保护与人工智能技术创新共存的全新秩序正在孕育萌生。

---

## 背景

生成式AI模型性能的飞跃，建立在对网络空间沉淀的大规模人类创作成果进行抓取（Scraping）并预训练的基础之上。然而，在这一过程中省略了创作者的明示同意或合理对价支付，不可避免地引发了接踵而至的法律纠纷。如果说早期的争论主要停留在是否构成未经授权的复制，那么如今已深化为一个更具根本性的结构性命题：AI模型的训练行为究竟是威胁现有创作生态的“市场替代”（Market Substitution），还是赋予原作品全新价值的“转换性使用”（Transformative Use）？

在此背景下，美国与欧洲探索了截然不同的应对路径。美国倾向于通过司法体系的事后判例寻求灵活的法理阐释，而欧盟则选择通过事前立法，强制要求提升透明度并保障权利人的保留权利（Opt-out）机制。这两种路径不仅深刻影响着全球科技巨头的AI研发方式，也从根本上重塑了数字内容出版商、媒体以及创作者的生存法则。

---

## 核心焦点

当前全球AI版权争议的核心焦点，主要围绕美国司法裁决与欧盟制度化立法体系这两大主轴展开。

### 1. 美国法院的“合理使用”审理与平台责任的边界

在美国法庭上，联邦《版权法》第107条所规定的“合理使用”（Fair Use）适用范围成为最大的角力场。在《纽约时报》（NYT）诉微软（MS）和OpenAI的侵权案中，联邦地方法院驳回了被告对核心诉求的撤诉申请，决定进入实体审理。这明确表明，法院并未一概接纳科技巨头关于AI训练属于合理使用的主张，而是决意严格审视其实质性侵权问题。反之，2026年9月，美国司法部（DOJ）提交了一份正式法庭陈述（Brief），倾向于支持OpenAI，认为AI模型对训练数据的使用可构成合理使用，这引发了行政当局与媒体、创作者阵营之间的剑拔弩张。

与此同时，司法机构也对无节制泛化的侵权指控划定了防线。美国联邦第九巡回上诉法院在涉及OpenAI、微软及GitHub的诉讼中，驳回了原告试图过度扩大《数字千年版权法》（DMCA）第1202条（版权管理信息）违规责任的主张。此外，在针对Meta的集体诉讼中，法院也因原告未能证明实质性的市场价值贬损（Market Dilution）而驳回了其请求。

然而，证据开示（Discovery）程序中曝光的内部文件在法庭外掀起了轩然大波。解密文件显示，微软某高管在内部将LLM的内容抓取定性为“人类历史上最大规模的劳动盗窃”（the largest labor theft in human history），这让外界对科技巨头在技术光环背后心知肚明其结构性侵权问题的质疑愈演愈烈。

### 2. 欧盟《人工智能法案》与TDM监管的确立

与美国以判例为主的事后规制形成鲜明对比的是，欧盟通过将《人工智能法案》（EU AI Act）与《数字单一市场版权指令》（CDSM）相结合，确立了明文化的规则。欧盟委员会为落实AI Act第53条及第55条的实践执行，出台了《通用人工智能实践守则》（GPAI Code of Practice）。

该监管框架依据算力规模对透明度义务进行了梯次划分：
* **计算量达到或超过 $10^{23}$ FLOPs 的模型**：必须公开详细的训练数据摘要，并遵守欧盟版权规则。
* **计算量超过 $10^{25}$ FLOPs 的模型**：被归类为存在“系统性风险”（Systemic Risk）的模型，须接受更为严苛的技术审查与治理监管。

尤其引人注目的是文本与数据挖掘（TDM）例外条款。即使是开源AI模型，若被认定具有系统性风险，亦不得享有豁免；且所有主体均必须遵守CDSM指令第4条第3款规定的版权方机器可读之“保留权利”（Opt-out）机制。对此，OpenAI、谷歌、微软等表态愿意遵守实践守则，但Meta及部分中国主流AI开发商拒绝签署，凸显出围绕规则认同的全球AI市场正在走向分化。

---

## 多维分析

法院与监管机构设立的基准，对技术基础设施和数据经济的运转机制产生了直接的连锁反应。这需要跳出单纯的法理争辩，从商业模式、网络基础设施及产业竞争力等多个维度进行深入审视。

### 1. 商业补偿与数据许可模式的重塑

随着法律风险的具象化，科技巨头为了应对诉讼并确保高质量数据的稳定供应，开始试验新型补偿模式。谷歌推出了“AI贡献试点”（AI Contribution Pilot）以及按价值付费（Pay-per-value）的许可模式，根据内容在生成式搜索（AI Overviews与Gemini）生成答案中的贡献程度向出版商分配收益。OpenAI也在ChatGPT中测试基于超链接的按点击付费（CPC）模式，探索针对面临流量下滑困境的新闻机构及内容创作者的实质性补偿方案。

这标志着行业正在摆脱过去搜索引擎单纯通过提供链接与网络生态共存的旧模式，转向在AI直接提供完整答案的“零点击”（Zero-click）环境中，向数据原创者直接回馈价值的新型数据交易模式。

### 2. 基础设施层面的网络控制技术普及

用于自动化合规并切实保护创作者权益的网络基础设施技术也在迅速升级。例如，Cloudflare开始推广“Accountable”与“Bot Preference Sync”等功能，使网站在保持常规搜索引擎正常收录的同时，能够精准定向拦截仅用于AI模型训练的未授权爬虫。

这突破了以往依赖静态、被动的 `robots.txt` 文件的局限，表明内容发布者获得了更细颗粒度的控制权（Granular Control）与技术主权，能够自主决定其作品究竟是仅用于搜索索引，还是同时可作为AI训练数据。

### 3. 对过度监管与市场固化的反思

然而，对于加强监管和征收许可费用可能带来的负面效应，业界也存在不容忽视的担忧。过高的数据补偿诉求和僵化的退出（Opt-out）监管，被普遍认为是导致先进基座模型在强监管地区推迟上线的重要诱因。

更有产业层面的警告持续指出，当只有资金雄厚的少数科技巨头能够承担天文数字般的数据许可费用与合规成本时，新兴AI初创企业的进入壁垒将被大幅推高，最终可能导致少数巨头主导的市场寡头垄断局面进一步固化。

---

## 展望

生成式AI与版权的关系，已明确从“无节制掠夺训练数据”阶段迈入“制度化交易与规范化审查”阶段。未来AI生态的演进将取决于以下三个关键分水岭：

第一，美国司法体系的终审定调与“合理使用”新边界的确立。《纽约时报》诉OpenAI案的实质审理，将成为权衡AI训练的“转换性价值”与“市场替代效应”真实权重的司法试金石。不过，结合美国司法部支持合理使用的意见及上诉法院的判例趋势来看，美国大概率将继续延续在不扼杀技术创新活力的前提下，合理平衡与界定平台责任的总体倾向。

第二，全球监管的两极分化与市场割裂。尽管欧盟《人工智能法案》与实践守则（GPAI CoP）设定的算力门槛（$10^{23}$ 及 $10^{25}$ FLOPs）和TDM退出机制旨在树立全球标准，但从Meta和中国开发者的拒签举态可以看出，“顺应监管”与“规避监管”两大阵营之间的技术差距以及区域性服务投放的差异化正在加剧。

第三，基于价值的数据经济能否真正确立。如果谷歌的AI贡献试点、OpenAI的CPC计费模式以及网络层面的精细化拦截工具能够在市场中稳固扎根，网络数据将不再是被免费开采的原料，而是演变为具有明确价格标签的数字资产参与流转。

人工智能时代的知识产权秩序，是一场在创作者合理回报与人类共同技术进步这两大价值之间探寻精密平衡的过程。法律与基础设施能否跟上技术的演进步伐、建立全新的共存规则，将成为决定未来数十年数字文明生产力高度的分水岭。

## Claims

- EU AI Act 실천강령 하에서 10^23 FLOPs를 초과하는 연산량으로 훈련된 GPAI 모델은 투명성 및 저작권 정책 요건을 준수해야 한다. (Supported)

## Forecasts

- 75% — 미국 연방법원의 AI 훈련 데이터 1차 공정이용 판결 도출 (2027년 상반기). Signal: 뉴욕타임스 대 오픈AI 약식판결 심리 결과 및 미 법무부 의견서가 법원 판단에 반영되는지 여부
- 65% — 빅테크 기업의 EU GPAI 규제 대응에 따른 파운데이션 모델 역내 출시 지연 (2026년 하반기). Signal: Meta 등 비서명 기업의 EU AI Act 제53조 투명성 요건 불응에 따른 집행위원회 조사 착수 여부

## Sources

- [Earl Spencer defends Diana book claims about King Charles - BBC News](https://www.bbc.co.uk/news/articles/cmqxvd1drd35o?at_medium=RSS&at_campaign=rss) — bbc.co.uk, 2026-09-19
- [Flight chaos caused by millisecond software defect, says air traffic control body - BBC News](https://www.bbc.co.uk/news/articles/cw0kl1571lpmo?at_medium=RSS&at_campaign=rss) — bbc.co.uk, 2026-09-19
- [Billionaire Manchester United owner Sir Jim Ratcliffe says he has lost confidence in UK - BBC News](https://www.bbc.co.uk/news/articles/cm0463619r1no?at_medium=RSS&at_campaign=rss) — bbc.co.uk, 2026-09-19
- [Google's Gemini AI hacked three companies in security test - BBC News](https://www.bbc.co.uk/news/articles/c607l0k72rlvo?at_medium=RSS&at_campaign=rss) — bbc.co.uk, 2026-09-19
- [All smiles in Strasbourg but uncertainty clouds Canada’s EU membership plan | European Union | The Guardian](https://www.theguardian.com/world/2026/sep/17/smiles-strasbourg-uncertainty-canada-eu-membership-plan-mark-carney) — theguardian.com, 2026-09-19
- [AI타임스](https://www.aitimes.com/) — aitimes.com, 2026-09-19
- [News and Analysis on Cryptocurrencies, Blockchain and Decentralized Finance - Cryptonomist](https://en.cryptonomist.ch/) — en.cryptonomist.ch, 2026-09-19
- [IPDaily | 지식재산 전문 미디어](https://www.ipdaily.co.kr/) — ipdaily.co.kr, 2026-09-19
- [Home - AI Insider](https://theaiinsider.tech/) — theaiinsider.tech, 2026-09-19
- [Generative AI - Wikipedia](https://en.wikipedia.org/wiki/Generative_AI) — en.wikipedia.org, 2026-09-19
- [OpenAI - Wikipedia](https://en.wikipedia.org/wiki/OpenAI) — en.wikipedia.org, 2026-09-19
- [Clearview AI - Wikipedia](https://en.wikipedia.org/wiki/Clearview_AI) — en.wikipedia.org, 2026-09-19
- [Trained AI models are *arguably* fair use under the "transformative" prong (whic... | Hacker News](https://news.ycombinator.com/item?id=33276345) — news.ycombinator.com, 2026-09-19
- [> Any time you use something with "fair use" in mind, it is the equivalent of sa... | Hacker News](https://news.ycombinator.com/item?id=39075282) — news.ycombinator.com, 2026-09-19
- [This analogy doesn't work. Fair use is an affirmative defense to copyright infri... | Hacker News](https://news.ycombinator.com/item?id=38815958) — news.ycombinator.com, 2026-09-19
- [> isn't this a slam dunk case? Meta literally published a paper where they said ... | Hacker News](https://news.ycombinator.com/item?id=42974403) — news.ycombinator.com, 2026-09-19
- [한국데이터경제신문](https://www.dataeconomy.co.kr/) — dataeconomy.co.kr, 2026-09-19
- [Davis Wright Tremaine](https://www.dwt.com/) — dwt.com, 2026-09-19
- [The New York Times v. Microsoft and OpenAI - Wikipedia](https://en.wikipedia.org/wiki/The_New_York_Times_v._Microsoft_and_OpenAI) — en.wikipedia.org, 2026-09-19
- [Fair use - Wikipedia](https://en.wikipedia.org/wiki/Fair_use) — en.wikipedia.org, 2026-09-19
- [OpenAI | Research & Deployment](https://openai.com/) — openai.com, 2026-09-19
- [General-Purpose AI Code of Practice - Wikipedia](https://en.wikipedia.org/wiki/General-Purpose_AI_Code_of_Practice) — en.wikipedia.org, 2026-09-19
- [> Nvidia wouldn’t say where this training data came from, but at least one repor... | Hacker News](https://news.ycombinator.com/item?id=42644416) — news.ycombinator.com, 2026-09-19
- [> Also explains why Apple isn’t launching in EU I don't buy it. From the AI act:... | Hacker News](https://news.ycombinator.com/item?id=41003909) — news.ycombinator.com, 2026-09-19
- [> the EU still doesn't have a clear definition of open source AI One can debate ... | Hacker News](https://news.ycombinator.com/item?id=41957772) — news.ycombinator.com, 2026-09-19
- [The New York Times Amends Lawsuit Against OpenAI and Microsoft - The New York Times](https://www.nytimes.com/) — The New York Times, 2026-09-19
- [Getty Images largely loses landmark UK lawsuit over AI image generator - Reuters](https://www.reuters.com/) — Reuters, 2026-09-19
- [Exclusive | Meta Approaches Media Companies About AI Content-Licensing Deals - WSJ](https://www.wsj.com/) — WSJ, 2026-09-19
- ["초5인데 국가대표라고?" 11살 日 게임 천재, 'AI급' 실력으로 金 노린다..."메달 따면 AG 역대 최연소 대기록"](https://www.chosun.com/sports/sports_general/2026/09/19/HAYTQZRWMJQTOZLEGBSGGNZWGI/) — chosun.com, 2026-09-19
- [When Trump and Xi meet they will discuss AI. 'Track Two' talks are already buzzing](https://www.npr.org/2026/09/18/nx-s1-5971481/trump-xi-meeting-ai-track-two-talks) — npr.org, 2026-09-19
- [Will even one of the U.N.'s 17 'sustainable development goals' be met by 2030?](https://www.npr.org/2026/09/18/g-s1-143738/united-nations-sustainable-development-goals-hunger-climate-gender) — npr.org, 2026-09-19
- [OpenAI Ordered to Hand Over 20M ChatGPT Logs in NYT Copyright Case - Decrypt](https://decrypt.co/) — Decrypt, 2026-09-19
- [News generative AI deals revealed: Who is suing, who is signing? - Press Gazette](https://pressgazette.co.uk/) — Press Gazette, 2026-09-19
- [California District Court upholds transparency requirements for generative AI training data - Norton Rose Fulbright](https://www.nortonrosefulbright.com/) — Norton Rose Fulbright, 2026-09-19
- [How Big AI Developers are Skirting a Mandate for Training Data Transparency - Tech Policy Press](https://techpolicy.press/) — Tech Policy Press, 2026-09-19
- [Researchers have trouble finding AI training data summaries - euractiv.com](https://www.euractiv.com/) — euractiv.com, 2026-09-19
- [Historic NYT v. OpenAI copyright battle heats up - axios.com](https://www.axios.com/) — axios.com, 2026-09-19
- [AI Companies Prevail in Path-Breaking Decisions on Fair Use - crowell.com](https://www.crowell.com/) — crowell.com, 2026-09-19
- [Two California district judges rule that using books to train AI is fair use - whitecase.com](https://www.whitecase.com/) — whitecase.com, 2026-09-19
- [Britannica Sues OpenAI Over ChatGPT 'Memorizing' Content - The Tech Buzz](https://www.techbuzz.ai/) — The Tech Buzz, 2026-09-19
- [Rightsholders Take the Lead: GEMA v. OpenAI - Taylor Wessing](https://www.taylorwessing.com/) — Taylor Wessing, 2026-09-19
- [Google Strikes Tough Negotiating Stance With Publishers on AI Licensing - The Information](https://www.theinformation.com/) — The Information, 2026-09-19
- [The price of AI training data, from $5M to $250M - qz.com](https://qz.com/) — qz.com, 2026-09-19
- [Here are the companies OpenAI has made deals with to train ChatGPT - Fast Company](https://www.fastcompany.com/) — Fast Company, 2026-09-19
- ["AI학습데이터 활용 시의 저작권, 법적 불확실성 해결해야" - 대학지성 In&Out](https://www.unipress.co.kr/) — 대학지성 In&Out, 2026-09-19
- [온신협·신문협 '생성형 AI 뉴스 저작권 침해' 공동 대응 - 한국기자협회](http://m.journalist.or.kr/) — 한국기자협회, 2026-09-19
- [AI로 교과서 학습해 문제집 팔면 저작권 침해…정부, AI 학습 가이드라인 발표 - 조선일보](https://www.chosun.com/) — 조선일보, 2026-09-19
- [DOJ urges judge to rule for OpenAI, Microsoft in N.Y. Times lawsuit - The Washington Post](https://www.washingtonpost.com/) — The Washington Post, 2026-09-19
- [Stability AI challenges Getty copyright claim over watermark intent standard - Daily Journal](https://www.dailyjournal.com/) — Daily Journal, 2026-09-19
- [Getty Images vs. Stability AI copyright infringement lawsuit](https://copyrightlately.com/pdfviewer/getty-images-v-stability-ai-complaint/) — copyrightlately.com, 2026-09-19
- [OpenAI strikes ‘first of its kind’ deal with publisher Axel Springer to use its news content in ChatGPT | CNN Business - CNN](https://www.cnn.com/) — CNN, 2026-09-19
- [[AI 시대의 저작권] ③ 학습 데이터 전쟁 격화…창작자 보상 시스템 시급 - 매일경제 마켓](https://stock.mk.co.kr/) — 매일경제 마켓, 2026-09-19
- [학습하고 싶은 'AI' VS 거부하는 '창작자'[AI시대와 저작권①] - 네이트](https://news.nate.com/) — 네이트, 2026-09-19
- [인공지능 학습을 위한 저작권법 개정이라니 [왜냐면] - 한겨레](https://www.hani.co.kr/) — 한겨레, 2026-09-19
- [What are the copyright challenges of AI training? - coe.int](https://www.coe.int/) — coe.int, 2026-09-19
- [European Parliament Proposes Changes to Copyright Protection in the Age of Generative AI - Global Policy Watch](https://www.globalpolicywatch.com/) — Global Policy Watch, 2026-09-19