Your AI content just got a name tag. Here's what that means for you

Anthropic confirmed on Tuesday 11th August 2026 that Claude will now watermark the text it generates. Not visibly, you won't see a stamp on the page but embedded in a way that's designed to survive copy and paste, and in some cases, survive editing too. It's part of Anthropic's response to the EU AI Act's transparency rules, which became applicable on 2 August 2026. Anthropic isn't alone: Google, Meta, Microsoft and OpenAI have all signed up to versions of the same commitment.

Sound like something that only affects tech companies? It doesn't. If you or your team use AI to write, design, or produce anything that goes out under your brand's name, this blog’s worth ten minutes of your time and a few link clicks.

What's actually changed

A quick timeline, because the dates get muddled easily. The EU AI Act itself entered into force on 1 August 2024, after being published in the Official Journal which is when it became law, not when most of it started applying. Article 50, which sets out the transparency obligations for generative AI, only became applicable on 2 August 2026, giving providers and businesses two years to get ready. We aways need a deadline to put something into action don’t we? The Commission backed a voluntary Code of Practice to help providers meet that deadline without each one inventing their own approach, and Anthropic confirmed it had signed on and rolled out Claude's watermarking on 11 August 2026, just a few short days after the deadline landed. Most of the industry has followed suit.

In practice, for Claude, that means two things. Any Claude model released after 2 August 2026 automatically embeds an invisible watermark in the text it produces, across Claude, Claude Code, Claude Cowork, Claude Tag and the API. Older models are being retrofitted to do the same. Files, images, SVGs, etc, are destined to get something different: signed provenance metadata using the C2PA standard, the same approach camera manufacturers and news organisations use to prove where an image came from.

The watermark travels with the text when it's copied and pasted. It may survive a light edit. Anthropic hasn't said exactly how much rewriting or paraphrasing is needed to strip it out, and to be fair to them, nobody in the industry has published a clean answer to that yet either.

Does this affect you? Ask yourself three questions.

1. Do you sell into, market to, or publish content for an EU audience, even occasionally?
2. Does your team use generative AI: Claude, ChatGPT, Gemini, Copilot, any of them to draft or produce anything that goes public? And
3. does any of that content touch matters of public interest, or feature real people, with little to no human review before it's published?

If you can answer Yes to any of those, this is relevant to how you work, even from the UK. The Act is EU law, but like GDPR before it, its reach follows the audience, not the postcode.

A quick checklist

Before you file this under "someone else's problem," run through this:

• Map which AI tools you or your team actually uses to produce public-facing content, and check whether the provider has signed the Code of Practice (Anthropic, Google, Meta, Microsoft and OpenAI all have).
• Decide together when you'll disclose that something is AI-generated or AI-assisted, particularly for anything touching news, current affairs, or public interest topics that hasn't had a proper editorial pass.
• Keep a genuine human-review step in your content process. It's good practice regardless, and under the Code, editorial review is one of the things that can exempt you from certain disclosure duties.
• Don't strip watermarks or provenance metadata deliberately. That's a different, more serious conversation than simply not knowing they're there.
• Check your AI vendor contracts and terms for what they've actually committed to on marking and labelling, don't assume it's the same across every tool you use.
• Keep an eye on guidance from the EU AI Office as enforcement detail firms up. It's still early days.

Where to go for more information

Two places worth bookmarking, rather than relying on second hand summaries (including this one): the European Commission's guidelines on AI transparency obligations, and the Code of Practice on Transparency of AI-Generated Content itself, which sets out what providers and deployers are actually signing up to. If any of this touches a live decision for your business, a contract, a compliance question, or a client commitment for example, that's a conversation for a solicitor who specialises in AI or data regulation, not a blog post. This one's here to help you work out whether you need that conversation at all.

What this means for students

If you're a student using Claude to help with an assignment, this is worth knowing: the text it generates may carry an invisible watermark that survives copy and paste, and possibly some editing. Universities are increasingly alert to AI-assisted work, and detection tools built around watermarking and provenance standards are only going to get more common in academic integrity checks.

That doesn't make it foolproof. Heavy rewriting or paraphrasing can still break the watermark, and Anthropic hasn't confirmed how much editing that takes. So it's not a green light to submit AI text unedited and assume it'll pass undetected, and a paraphrased version isn't automatically "safe" either. The honest answer is the same one I'd give any student asking about AI and assessment: use it as a sparring partner for thinking, not a replacement for doing the thinking. That was true before the watermark. The watermark just makes the shortcut version a worse bet than it used to be. I encourage the use of AI tools in the research part to help speed up this once lengthy process, but you cannot speed up the human brain’s ability to read this back to you and digest, this still takes processing time (and often coffee I hear) for many. As a student, your personality needs to shine through the words so they sound like you do in class, the tutors are aware of this; ever wondered why they encourage you to speak in class?

What this means for marketing teams

For anyone producing content at pace; blog posts, social copy, ad variations, email drafts you get the idea, this shifts the question from "are we allowed to use AI?" to "have we thought about disclosure and review?" If your team uses AI-generated text or imagery anywhere public-facing, especially anything that reads as commentary on current affairs, or images of real-looking people or places, you need a house policy on when that gets flagged and when it gets a human pass before it goes live.

There's a trust angle here too, and it's the one I'd lead with. Clients and audiences are getting more attuned to what's real and what's generated. Being upfront about where AI sits in your process and having the governance to back it up is fast becoming a mark of a business that takes its content seriously, not a confession. That's the point of a well-governed content pipeline: it lets your team's actual expertise do the talking, instead of leaving anyone guessing what's underneath the copy.

How will this actually be checked?

Honestly, the enforcement mechanics are still taking shape. Article 50 compliance sits with national market surveillance authorities, the EU AI Office and the European Data Protection Supervisor, and the Code of Practice is the Commission's way of giving providers a clear, common standard to work to rather than each one guessing. What's less settled is exactly how detection gets applied day to day, whether that's platforms scanning for C2PA metadata on uploaded images, or watermark-detection tools built into publishing and social tools further down the line. If you want a definitive answer on how checks will run in your sector, that's genuinely one to watch for as the guidance matures rather than something anyone can tell you with confidence today. But be prepared the checks are coming and will of course probably have some sort of AI way of doing this quicky, so start getting into good habits and sparring with your AI not copying and pasting

If you want a hand working out where your team's AI use actually sits against all this, what needs a policy, what needs a human check, what's already fine we can support you and your teams, you know how to get in touch.

Useful Sources to bookmark:

• AI Act
• Article 50(2), (4) and (5) of the AI Act.
• Introducing Official Content Credentials Icon – C2PA
• EU Icons for Labelling AI-Generated Content – European Commission
• Commission publishes Code of Practice on marking and labelling AI-generated content
• The Code of Practice on transparency of AI-generated content (PDF)
• Guidelines on the scope of the transparency obligations laid down in Article 50 of the AI Act.
• ICO.org guidance on UK GDPR on AI and data protection
• GDPR Advisor – EU Artificial Intelligence Act: The Complete Compliance Guide for 2026
• ICO.org guidance on UK GDPR on AI and data protection
• GDPR Advisor – EU Artificial Intelligence Act: The Complete Compliance Guide for 2026