Anthropic has confirmed it will embed an invisible watermark in all text generated by Claude, its AI model. The move, announced this week, is designed to comply with Article 50(2) of the EU AI Act, which requires AI providers to make outputs detectable as machine-generated. The watermark will be applied globally, not just in the EU, and is designed to survive copying, pasting, and some editing.
The technical mechanism is likely statistical watermarking. As Claude predicts the next word in a sequence, it slightly increases the probability of choosing certain words over others. Over a long enough passage, that creates a detectable pattern unique to Claude's output. Anthropic says the mark will not confirm the full provenance of content, and heavy editing could remove it.
OpenAI, Google, and other major model providers are expected to follow with similar measures to comply with the same regulation.
What it means for legal AI platforms
Claude and other frontier models sit underneath many of the legal AI products now being deployed in law firms. That raises a practical question: will a contract drafted with AI assistance carry a watermark by the time it has been copied into Word, edited through several rounds of markup, and prepared for signature? The same applies to client memos and court filings.
Arnav Joshi, an AI and technology partner at Ashurst Perkins Coie, said the marking requirements have been expected by the industry since the AI Act came into force in August 2024. He said they are likely to apply both to model providers and to the platforms built on top of them.
"The fact that a legal AI platform is built on somebody else's model doesn't necessarily take it outside these requirements," Joshi said. "There are obligations that potentially run through the chain. From an AI Act perspective, how compliance obligations like watermarking are actually met will sometimes rely on close cooperation between model providers and companies integrating those models."
Harvey declined to comment on how it is approaching the transparency requirements. Legora did not respond by the time of publication.
AI use becomes visible
Most clients already expect their law firms to use AI, largely because they want firms to control costs. How work is completed has generally been left to the firm, provided the outcome is sound and the bill is reasonable. Watermarking could change that dynamic.
If AI-generated text carries a detectable marker as it moves from a legal AI platform into a document sent outside the firm, the specific uses of AI are no longer invisible. That feels relevant after several high-profile incidents this year. Pinsent Masons and Sullivan & Cromwell both faced episodes involving fabricated case authorities appearing in court documents. Those errors were caught because someone noticed the cited authorities did not exist. In a world of widespread watermarking, there may be another clue that AI was involved somewhere in the production of a document.
Joshi expects transparency around AI use to become standard practice, and he thinks that should broadly be welcomed as AI becomes an ordinary part of professional work.
"The EU has a track record of setting rules that end up becoming global standards - GDPR is the obvious example," Joshi said. "I think we could see something similar with the AI Act, where greater transparency around AI-generated content becomes the norm well beyond Europe. That shouldn't necessarily be viewed negatively. If AI is going to become a normal part of how professional work is produced, some transparency around its use makes sense."
The cost question
There is also a connection to how legal AI is priced. Major legal AI companies are moving toward consumption-based pricing, passing on some of the token costs they incur from model providers to their customers. Legora recently introduced consumption-based pricing for its premium agentic product, and Harvey has signaled it is heading in the same direction.
Law firms could start incurring material AI-related costs on individual matters, with some or all of those costs passed on to clients. Greater transparency around AI use could help firms justify those charges. If clients accept that AI was used in producing legal work, and that its use is disclosed and marked, it becomes easier for firms to explain why a matter incurred £500 or £50,000 of AI-related costs alongside lawyers' time. The same rules that make AI use more visible could make it easier to bill for.
Why this matters for legal professionals
The watermark changes what clients can observe about how their legal work was produced. Law firms that use AI-assisted tools should consider how they will respond when a client asks why a document carries an AI marker, or when opposing counsel raises it in litigation. The practical takeaway: document your AI use policies now, and be ready to explain how AI contributed to a piece of work before someone else asks first. For those looking to build familiarity with these tools, AI for Legal training covers the practical applications. Paralegals handling document review and contract analysis may find AI for Paralegals a useful starting point.
Your membership also unlocks:
