As you may have heard, Anthropic (Claude) has decided to generalise the watermarking of AI-generated text, going well beyond the simple framework set by the European Union. Other players have followed this trend, and it turns out that most have in fact been embedding watermarks in AI-generated text since 2024.
Our clients generally produce high-value writing for their principals, whether simple e-mails or complex reports. The temptation is strong to lean on AI to rework, or even fully draft, a text. To the writer, the practice seems harmless as long as their own thinking is faithfully conveyed.
The risk of detection and the loss of trust
Even though means of spotting artificial content already exist, the official tools to read these watermarks will soon be made available to the public by the tech giants. So whether you use Claude, Gemini, ChatGPT or others, the “AI signature” may be revealed to the recipient of your report or e-mail.
The direct consequence is that doubt sets in. Even if you wrote the document yourself and simply asked the AI to rephrase it, there is a good chance it will be flagged. It will then be impossible for the reader to know what percentage of the text was conceived or reworked by the machine. A loss of trust from the client can follow.
Whether you wrote it yourself or not, your text can be flagged. Doubt, however, cannot be watermarked away.
The challenge of internal control
Faced with this risk, some management teams now require texts to be 100% human-written. Yet it is very difficult to police this internally. We are already seeing, at some clients, amendments to employment contracts that simply prohibit the use of AI in client-facing external communications.
The solution: private AI
In this case as in others, true autonomy comes from deploying a private AI. You control the generation parameters (with or without watermarking) and, above all, your data stays strictly confidential: nothing leaves your premises.
This is not a minor technical detail: it is the difference between enduring a technology and mastering it. With an AI that belongs to you, the question of watermarking stops being a risk and becomes a simple setting: yours.