In brief: Since 2 August 2025, providers such as Anthropic and OpenAI have been required under the EU AI Act to label AI-generated content, with Anthropic now introducing invisible text watermarks for all future Claude models.
Since 2 August 2025, providers of particularly capable general-purpose generative AI models such as Anthropic and OpenAI have been required under the EU AI Act to ensure that users can recognise whether they are interacting with an AI and whether content is AI-generated. In response, Anthropic announced on Monday that it will equip all future Claude models with embedded text watermarks.
The transparency obligations of the EU AI Act have been binding since 2 August 2025 for providers of particularly capable generative AI models. Those affected include, among others, Anthropic and OpenAI. Anthropic announced that all Claude models released from this date onward will weave watermarks directly into the generated text. According to a company blog post, this watermark is invisible to users and does not alter the meaning, quality, or readability of the output. Because it is part of the text itself, it is said to be preserved even when copied and pasted into other contexts. OpenAI already uses SynthID, a technique developed by Google DeepMind for invisible overlays, for image content, and extended this to audio content at the end of July. Both companies also support the industry standard for content provenance labelling developed by the Coalition for Content Provenance and Authenticity (C2PA).
For compliance officers in enterprises, this step marks a transition from voluntary self-commitments to regulatorily enforced verification obligations. Until now, users themselves had to judge whether content was AI-generated – according to Ashley Casovan, Managing Director of the IAPP’s AI Governance Center, this is now shifting to a system of verified labels or overlays. Francesca Pole, a privacy and cybersecurity lawyer at DLA Piper, confirms that transparency obligations are currently at the centre of the discussion around implementation of the AI Act. For companies integrating generative AI models into their own products or workflows, this creates review obligations: it must be ensured that the models used meet the labelling requirements and that end users are informed accordingly.
This development confirms a recurring pattern: through the AI Act, the EU is de facto setting global standards, similar to how the GDPR previously did in the area of data protection. Providers such as Anthropic and OpenAI are implementing the technical requirements not only for the European market but across their models as a whole, effectively leading to a worldwide alignment with EU requirements. For compliance teams at DACH companies, this means that checking AI-generated content for provenance labelling is likely to become part of standard due diligence when procuring and deploying generative AI systems.
Source: www.politico.eu · Published 13 August 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.