Bottom line: Since August 2, 2025, labeling requirements for AI-generated content have applied under the EU AI Act, which Anthropic is meeting with embedded text watermarks in all new Claude models.
Since August 2, 2025, providers of particularly capable generative AI models have been required under the EU AI Act to ensure that users can recognize whether they are interacting with AI and whether content is AI-generated. In response, Anthropic has announced that it will equip all Claude models released since that date with embedded text watermarks.
The EU AI Act has obligated providers of advanced generative AI systems to establish transparency around AI-generated content since August 2, 2025. Users should be able to recognize whether they are interacting with a bot and whether a piece of text, image, video, or audio content was generated by AI. Anthropic announced on Monday that all Claude models released since that date feature a watermark embedded directly in the text. According to the company, this watermark is not visible to users, does not alter the meaning, quality, or readability of the response, and is preserved when the text is copied elsewhere, since it is an integral part of the text itself. OpenAI had already announced in late July that it would use SynthID, a watermarking tool for images developed by Google DeepMind that has since been extended to audio content as well. Both companies also support the industry standard for content provenance labeling developed by the Coalition for Content Provenance and Authenticity (C2PA).
For compliance officers at companies using generative AI models from providers such as Anthropic or OpenAI, or further processing their output, this marks a transition from voluntary commitments to a regulatorily enforced labeling obligation. Ashley Casovan, Managing Director of the IAPP’s AI Governance Center, describes the shift as a move away from a world in which users had to make an “informed decision” about the origin of content, toward a world with verified labels or overlays. Francesca Pole, a data protection and cybersecurity lawyer at DLA Piper, currently describes transparency obligations as the central topic of discussion in the context of the AI Act.
For practical implementation within companies, this means that internal policies governing the use of generative AI should be reviewed, particularly with regard to the further processing of AI-generated content in customer communications, marketing, or documentation. Since watermarks such as Anthropic’s are technically embedded in the text and preserved during copying, downstream verification mechanisms for establishing provenance may become more important — for example, when determining whether published content must be disclosed. The article also notes that with this regulation, the EU is once again taking a pioneering role in the global enforcement of platform standards, comparable to the impact of the GDPR on data protection worldwide.
Source: www.politico.eu · Published August 13, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrasing and classification by Lumi News Pipeline v1.8.3.