Updated September 8, 2026. When we published this we said Anthropic marks its text after signing Europe's transparency code. What we did not say, because in August we had not verified it, is what that rule actually requires, of whom, and since when. That is where the problem is, because what is being repeated says close to the opposite of what the text says. We added it below.
Direct answer: three things happened in AI this week that, together, tell one story. Anthropic started putting a hidden mark on everything Claude writes (Anthropic); there was a shake-up at the top of Google DeepMind (Axios, Aug 5); and Europol had already projected that by 2026 up to 90% of online content could be AI-generated (Europol Innovation Lab, 2022). The story underneath all of it: when anyone can produce correct-looking content with one click, the scarce and expensive thing becomes what is authentic. Here is what it means for you as a practice owner, in 4 minutes.
The big one: Claude now signs what it writes
Anthropic, the company behind Claude, started weaving an invisible mark into the text its AI generates. For files and images it also adds an origin label using an open standard called C2PA. It did this after signing the transparency code of Europe's new AI law, and it applies worldwide, not only there (Anthropic, Aug 2026).
Here is the nuance almost nobody underlines, and Anthropic says it plainly: the mark does not prove who wrote the text. It can disappear if you edit heavily, paraphrase, or translate, and detecting it does not confirm where the content really came from. In other words, it is not a foolproof detector of content made by AI. It is just a label that says "an AI passed through here at some point."
What it means for you. Many people will run to hide that they use AI. That is the wrong move. If even the company adding the mark admits it does not prove authorship, the lesson is the opposite: using AI to draft is fine, but what sets you apart is what you put on top of it. Your judgment, your experience, your real client story, the number you can actually back up. AI drafts; you supply the authorship and the trust. And now there is one more reason: hiding it is not what the rule asks for, and reviewing and signing what you publish is literally its exception.
The rule behind it, and it says close to the opposite of what you are reading
Above we said Anthropic marks its text after signing Europe's transparency code. The half that matters is missing: what that rule requires, of whom, and since when.
It is Article 50 of the European Union AI Act, and it came into force on August 2, 2026. Paragraph 2 says that whoever builds the system has to mark what that system generates, in a machine-readable format, detectable as artificially generated. That is exactly what Anthropic did, and it is why it did it in August.
Here is the part being reported wrong. The headline going around is "you can't use AI to write anymore." That is not what the rule says. The duty to mark a model's output belongs to whoever builds the model, not to you.
Now the honest nuance, because the article has more than one audience: paragraph 4 does put duties on whoever uses the system. If you generate text with AI and publish it for the purpose of informing the public on matters of public interest, you have to disclose that it was artificially generated. That sounds like it catches you, until you read the exception in the very same paragraph: the duty does not apply where the content has undergone human review or editorial control and a person or company holds editorial responsibility for publishing it.
Read that again, because that is the news. The rule does not reward not using AI. It rewards reviewing it and signing it. AI-assisted text that you read, correct and publish under your own name falls outside that duty because somebody is accountable for it, and that somebody is you.
What did change is smaller and more uncomfortable: what ended is not using AI, it is being able to pretend you didn't. The tool leaves a trace by design, and that changes where you stand, not what you are allowed to do.
And one honest clarification, because you're owed one: we know of no demonstrated ranking penalty for simply using AI to draft, and Google has said in writing that what it cares about is not how a text was produced. What is in its published policies, and worth not confusing, is content produced at scale to manipulate rankings, whether a person or a machine writes it. So: a watermark will not sink you. Publishing fifty generic pieces a month can cost you, and that was true before there was a law.
And since we are not lawyers and this is a rule with several audiences: if you publish from Europe, sell into Europe, or work in a regulated profession, this does not replace legal advice. We say it because it costs us one line and it can save you a problem.
None of this changes what was already true before there was a law, which is exactly what comes next. And where all of it sits in the full picture of your visibility is in the five disciplines map.
The number that changes everything
Before cheap AI existed, producing content had a cost. That cost was, without us noticing, a quality filter. That filter is gone. And this is not marketing hyperbole: Europol projected that by 2026, up to 90% of online content could be AI-generated (Europol Innovation Lab, 2022). It is an expert projection, not an exact measurement, but we are in that very year and the direction is unmistakable.
If nine out of ten pieces can be AI, what becomes valuable? What a machine cannot manufacture: a real person, with a name, judgment, and proof. Authentic stopped being the pretty thing and became the scarce thing.
Also this week
A shake-up at the top of Google DeepMind. Demis Hassabis steps aside as CEO of Google DeepMind to chair the unit and take on the role of chief scientist at Alphabet; and Jeff Dean, one of Google's landmark figures, is leaving the company after 27 years (Axios and CNBC, Aug 5 2026). For you this is not corporate gossip, it is a signal: the field moves so fast that even its founders rotate. Betting on your own brand and your own processes is steadier than marrying a single tool or a single name.
Anthropic wants to make Claude's engine cheaper. It is building a team to design its own custom chips (Quartz, Aug 5 2026). It is an intention, not something already on the street, but it points the same way: quality AI is going to cost less and less. When price stops being the barrier, having AI does not set you apart from anyone. What you do with it does.
Hub365 AI's take
Line up all three: AI signs the content, almost everything can be AI, and using it gets cheaper. It all pushes the same way. Volume is worth nothing now, because anyone can produce it.
So your edge as a professional moves to what is scarce: being real, verifiable, and citable. And this connects straight to how a client picks you today. 37% of searches already start on AI, and that number is the floor, not the ceiling, it keeps climbing (Eight Oh Two, 2026). When someone asks ChatGPT for the best in your specialty, AI names whoever it can understand and cite with confidence. Being that trusted source is exactly what a watermark cannot manufacture for you. We dig into it in AI search has no page two.
The Todd and Naty corner Todd: When producing gets free, the expensive thing is trust. The watermark does not take your credibility, it relocates it: now it lives in your proof, not your writing. Naty: It makes me laugh how everyone will read the watermark backwards. It is not "careful, they will catch you using AI." It is "generic no longer sets you apart, your voice does." Good news for anyone who actually has something to say.
What to do this week (2 minutes)
- Take one recent piece of yours (a post, an email to a patient or client) and add a layer AI cannot fake: one verifiable fact of yours, a real client situation with no names, an opinion you would defend.
- Check that your key information is still yours and citable: prices, area, specialty, in plain words AI can read.
- Do not run to hide that you use AI. Run to make your voice and your proof impossible to copy.
Get it every week before anyone else
This is the short version. Every Friday at 4 pm ET we send The AI Edge, our newsletter for practice owners: the week in AI translated into what it means for you, with the play and the prompt of the week. Subscribe free at hub365.ai and get it in your inbox before it publishes here.
