Your AI Text Is Now Watermarked: Article 50 of the EU AI Act for Developers
Canonical note: this is a shortened version. The full article, with sources and the downloadable checklist, was first published at https://bookpilot.org/blog/ai-act-article-50-watermarking-ai-text/ . General information only, not legal advice. If your app calls Claude or Grok, the text coming back carries, or will very soon carry, an invisible watermark. No code change, no new characters in the…
From 2 August 2026, the EU AI Act has required developers to watermarks AI-generated text in order to inform users that they are interacting with an artificial intelligence system. This obligation applies to any provider of generative AI systems, regardless of whether their models are already deployed or will be launched after the initial 2 August deadline.
Additionally, deployers of AI systems are also required to visibly disclose when AI-generated content is being shared, particularly if it concerns matters of public interest.
In response to these new regulations, major AI providers have announced their plans to implement watermarks on their systems. Anthropic, the creator of the Claude language model, has confirmed that all supported models will now include an invisible watermark across their API, apps, and Claude Code, globally. This change will take effect on 10 November.
Similarly, Amazon Bedrock's Grok models will begin watermarking text from 7 November 2026, including structured outputs and tool-call arguments. OpenAI has also hinted at a textGrain feature in its ChatGPT platform, which will be available only in Europe.
While the watermark itself is an invisible addition to the AI-generated text, providers must still ensure that users are made aware that they are communicating with an AI system. This disclosure can be achieved through various means, such as a clear notice stating "you're chatting with AI" within the chat interface. It is important for developers to note that the watermark provided by the AI provider does not substitute the legal requirement to disclose AI-generated content, especially when it involves matters of public interest.
In such cases, deployers must still visibly disclose the AI-generated nature of the content to their users.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
