Anthropic has begun deploying machine-readable watermarks for Claude, aligning with new European transparency laws and indicating a broader industry shift towards AI content verification standards.
Anthropic has begun rolling out machine-readable watermarks for text produced by Claude, a move aimed at making AI-generated material easier to identify as new European transparency rules take effect. According to TechCrunch and Axios, the system is designed so the mark can survive simple copy-and-paste use, although the company has not explained exactly how much editing would be needed before it disappears. Anthropic also says the feature will be added to future models automatically, with older systems to follow later.
The change reflects the European Union’s AI Act, whose transparency provisions entered into application on 2 August. The European Commission says the rules are intended to make it clear when people are dealing with AI content, including chatbots and deepfakes, and the Commission has published a voluntary code of practice to help providers and deployers meet those obligations. The legal framework also allows a later deadline for systems already on the market, giving providers until 2 December to retrofit older products.
Anthropic’s approach is not limited to Europe. Tom’s Hardware reported that the company intends to apply the watermarking globally and is preparing detection tools as well, while also acknowledging limitations. Those include the possibility that quoted material could be flagged incorrectly and the fact that unmarked text cannot automatically be assumed to be human-written. The company has also said image provenance will be handled through the C2PA standard, which is designed to attach verifiable origin data to media files.
The wider industry is moving in the same direction. Suno has said it will tag songs made on its platform after legal pressure, while Substack has worked with a detection company to flag AI-written posts after its chief executive criticised people for presenting machine-generated text as their own. Reuters has previously reported that Google, Meta, Microsoft and OpenAI have also agreed to similar transparency efforts, suggesting that watermarking may become a standard feature rather than a niche compliance tool.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
