Does copyright protect AI-generated content in the EU?

As of 2026, EU copyright law (InfoSoc Directive 2001/29/EC, as interpreted by CJEU) requires that protected works reflect the 'author's own intellectual creation', a standard that requires a human author making free creative choices (CJEU C-5/08 Infopaq, C-145/10 Painer). Fully autonomous AI-generated content, where no human makes substantive creative choices, does not qualify for copyright protection under current EU law. However, content produced with AI as a tool, where a human author directs, selects, and curates the output, may retain copyright to the extent the human's creative choices are expressed. The EU AI Act (Regulation 2024/1689) Art. 50 requires AI-generated content to be labelled, but does not resolve copyright ownership questions.

How does the EU AI Act affect copyright proof for AI-assisted content?

The EU AI Act (Regulation 2024/1689), fully applicable from August 2026, requires providers of general-purpose AI models (Art. 53) and systems generating synthetic content (Art. 50) to implement technical solutions ensuring AI-generated content is detectable. For copyright purposes, the AI Act introduces a duty to document training data (Art. 53(1)(d)) and to disclose that content is AI-generated. For content creators using AI: documenting your creative direction (prompts, iterations, editorial selections, modifications made to AI output) creates evidence of human creative input. A qualified electronic timestamp on your creative process documentation establishes when your creative choices were made, supporting any future copyright claim under the human-creativity standard.

Who owns copyright in content created with AI tools?

Copyright ownership in AI-assisted content varies by jurisdiction and creative process. In the EU/Switzerland: the human author who exercises creative choices in using the AI tool (prompt engineering, selection and curation, substantial editing) retains copyright to the extent their personality and creativity is expressed in the final work. The AI system has no legal personality and cannot own copyright. The AI company does not own copyright in user-generated outputs by default (though check specific terms of service). For maximum copyright protection of AI-assisted work: document your creative decisions and prompts, make substantial editorial modifications to AI output, and apply a qualified electronic timestamp to your final work and your process documentation. This establishes both the creation date and evidence of human creative involvement.

How can content creators protect their work from being used to train AI without consent?

Under EU law, the Text and Data Mining (TDM) exception (InfoSoc Directive Art. 4, amended by DSM Directive 2019/790 Art. 4) allows AI companies to scrape publicly available content for training unless the rights-holder has reserved this right using machine-readable means (e.g., the robots.txt ai-block directives or C2PA content credentials). Practical steps: (1) Add machine-readable AI opt-out signals to your website; (2) Apply C2PA content provenance metadata to your images and creative works; (3) Register with content provenance platforms; (4) Apply qualified electronic timestamps to your creative works. This establishes creation date and authorship for any future infringement claim if your work appears in training data without permission.

What evidence do I need to prove a human created AI-assisted content?

To prove human authorship in AI-assisted content: (1) Document the creative process by keeping your prompts, prompt iterations, the AI's raw outputs, and your editorial decisions; (2) Show substantial human modification, meaning where you rewrote, restructured, or substantially altered the AI output based on your creative judgment; (3) Apply a qualified electronic timestamp to the draft at different stages, showing the evolution from concept to final version proves that human creative choices drove the process; (4) Maintain version control of your work; (5) Create a project log explaining your creative vision and why you made specific choices. A qualified timestamp on your process documentation, not just the final work, is the strongest evidence because it proves your creative involvement existed at a specific time, before the final product was completed.

IP Copyright

2026年如何证明AI辅助内容的版权所有权

Dani Wattenhofer · Business Development, Swiss Trust Layer AG

2026年6月23日 7 分钟阅读最后更新 2026年9月8日

返回博客

In short

2025年3月，一家位于巴塞尔的用户体验机构在客户合同正式对签的三天前，向一家A轮融资初创企业交付了完整的品牌视觉识别方案。交付两周后，该初创公司的新任内部设计师宣称对整个视觉系统拥有署名权。该机构在整个项目中全程使用了人工智能工具：每个提示词都经过精心引导，每项输出都经过修改，每个字体决策都是深思熟虑的结果。但没有任何草稿、项目简报或迭代文件上存在合格的时间戳。初创公司的律师向机构法律顾问提出了唯一一个问题：请证明该机构是最先创作的一方。该机构没有任何能够在法庭上站得住脚的答复。

那场争议的结果，并不取决于实际完成工作的是谁，而是取决于谁在创作过程进行时留存了记录。欧盟一起普通知识产权侵权案件平均花费EUR 25万至EUR 120万（EUIPO，2023年）。该机构的项目简报、第一稿以及三轮修改，每一份都本可以在完成时被封存。四次封存所对应的，却是六位数的法律风险敞口。价格公布于 swisstrustlayer.com。

查看关于人工智能内容所有权的完整指南

使用 Swiss Trust Layer AG 保护您的作品

使用由 Swisscom Trust Services 支持的司法认可 e-Seal 封存您的知识产权。

预约免费演示

IP & Copyright

丢失所有密码之后：恢复靠的是你本人，而不是一次登录

任何服务都能给你发一封重置邮件。可一旦连那个邮箱、手机、密码管理器和所有备用验证码都丢了，它们就帮不上忙了。ESCROW Storage 与你经护照验证的身份绑定，所以取回已封存的文件，不取决于你手上还剩下哪个凭证：你可以亲自到场取回。

2026年9月30日阅读文章 →

IP & Copyright

云存储不是保管：瑞士身份核验托管存储究竟改变了什么

Dropbox、谷歌云端硬盘、公司共享文件夹：它们都把访问权交给持有登录信息的人，而不是文件真正的所有者。身份核验托管存储的运作方式不同，恢复权与经 KYC 核验的本人绑定，文件保存在瑞士，独立于任何云服务提供商。

2026年9月25日阅读文章 →

IP & Copyright

AI版权诉讼实际进展到哪一步了,又有哪些问题是它们谁都没替你回答

四起案件常被笼统地称为"AI版权诉讼"。它们并非同一起案件,而且没有一起就未经许可用你的作品训练模型是否构成侵权作出过裁决。以下是安德森案、纽约时报诉OpenAI案、盖蒂诉Stability案和巴茨诉Anthropic案的真实进展,以及无论这些案件如何或何时收场,创作者自己能够证明什么。

2026年9月22日阅读文章 →

IP & Copyright

工程师走出门口之前的那一刻,才是真正重要的时刻

知识产权风险不是从辞职信开始的,它集中在此前的几周内:没人标记为公司记录的私人代码仓库、个人邮件和未合并的分支。把源代码、设计文档和原型作为日常工作而非只在发布时封存,才能让初创公司的所有权在任何人离职时都经得起证明。

2026年9月21日阅读文章 →

IP & Copyright

OpenAI、Google 和 Nvidia 都支持内容凭证。这依然不等于法庭上的证据

内容凭证如今来自 ChatGPT 生成的图片、专业相机,很快还会来自 Chrome 本身。这让来源信息可以大规模被读取。但它并不能让一份清单变成法院推定其日期准确的东西。

2026年9月17日阅读文章 →

相关文章