The AI Act and Watermarks: A Practical Guide for Entrepreneurs and Businesses
Tutorial

The AI Act and Watermarks: A Practical Guide for Entrepreneurs and Businesses

Find out what will change starting August 2, 2026, with the AI Act. Learn how to manage AI-generated content in the office while ensuring transparency and security.

Redazione Leader24August 13, 20266 min readSpunto da TechCrunch AI

You’ve just finished writing a proposal for a new client. It’s a great proposal—clear and compelling—and you completed it in half an hour instead of two days, thanks to an AI assistant. You send it off without a hitch. Then you discover that as of August 2, 2026, the rules have changed: all AI-generated content must be identifiable and marked with a watermark. You wonder if you’ve done something wrong or if you’re breaking the law. The short answer is probably not, but you need to understand what has actually changed. Watermarking technology is still in its infancy, and the responsibility ultimately remains yours.

What Are Watermarks and Why Does the AI Act Make Them Mandatory?

Watermarks are invisible digital signatures embedded in a file to indicate that a particular text, image, or audio clip was created by artificial intelligence. The European AI Act has made this labeling mandatory to ensure transparency. The goal is for anyone reading your document to know whether a person or a machine is behind it. Article 50 of the AI Act, as explained in this analysis by Lex Machina, makes no distinction between types of content: text, images, video, and audio must all be marked. But the real question is this: does this watermarking truly protect your business, or does it just give you a false sense of security?

Does the watermark really protect my privacy or my brand?

No, the watermark is not a cybersecurity tool. It doesn’t protect your brand from imitations, it doesn’t block fake profiles, and it doesn’t defend against attacks. An entrepreneur asked me if adding a watermark prevents someone from copying his content, and the answer is no. It only serves to indicate that the material was generated by AI—not to prevent it from being stolen or modified. The confusion is widespread because many people think that transparency is the same as protection. If a malicious actor creates a fake profile for your company and publishes content, the watermark won’t stop them. Brand security is a different ballgame, one that relies on monitoring tools and constant human oversight.

Why Watermarking Technology Is Still Vulnerable

This is the sticking point. Watermarks on text are easy to remove: even a slight paraphrase or rewrite is enough to make the digital mark disappear. An in-depth analysis by Cybersecurity360 identifies three structural limitations: dependence on text length, vulnerability to scrubbing attacks, and the risk of false positives. Relying solely on technology to ensure compliance is dangerous. If the watermark disappears, legal liability for potentially misleading content falls on whoever published or sent it—that is, you.

How to Manage AI in the Office Without Legal Risks

The key isn’t the watermarking technology itself, but human oversight. In practice, this means three things. First, every AI-generated text must be reviewed and approved by a real person, without exception. Second, if you use AI to assist customers, state this clearly. Finally, if you use an automated system to respond on WhatsApp or your website, make sure you can step in when needed. Platforms like Leader24 allow you to switch from AI to a human agent in real time, while maintaining control over the conversation.

What Tools to Use to Avoid Getting Overwhelmed by Obligations and Management

You don’t need a separate software solution for every task. The ideal approach is to centralize activities that require customer contact. You can use Calendly to schedule appointments without endless email exchanges, Trello to organize team tasks and track who reviewed what, or the platform if you want to combine the speed of AI to respond to customers on WhatsApp or your website, while ensuring a human can step in at any time to correct or handle complex cases. The key is not to delegate to AI what requires human judgment. Use AI to speed things up, but stay in the driver’s seat.

What Happens If a Customer Feels Misled by AI

Reputational damage is more serious than an administrative fine. According to Nova IUS, the careless use of AI creates new legal and reputational risks. A customer who discovers they’ve been talking to a chatbot without realizing it feels deceived, and a customer who feels deceived won’t come back. Be honest: if a customer messages you, they’d rather have a clear, human response than an automated reply that sounds fake. Transparency isn’t a cost—it’s a competitive advantage.

Frequently Asked Questions

Do I have to put a watermark on every text I write with AI?

No, you don’t have to do that yourself. The labeling requirement falls on AI model providers such as OpenAI, Google, and Anthropic. Your job is to ensure that the content you publish or submit is identifiable as AI-generated, if it is. The technical labeling is handled by the software.

If I rewrite an AI-generated text, do I still have to disclose it?

If you completely rewrite it with substantial human input, you are no longer required to disclose it. However, if you retain the structure and main ideas, transparency remains a best practice. When in doubt, state that you used AI as an assistant.

What are the risks if

I don’t comply with the transparency requirement?

The penalties under the AI Act can amount to a significant percentage of global annual revenue for companies that violate the regulations. For a small business, however, the greatest risk is damage to its reputation. A customer who feels deceived is unlikely to give you a second chance.

The first step to take today is an internal audit. Make a list of where you use AI: email, chat, social media, reports. For each item, ask yourself whether that content, when read by your best customer, would reveal that it was written by an AI. If the answer is no, add a transparency note or include a human review step. Technology is an excellent assistant, but the signature and responsibility must always remain yours.

Leader24 Insights

If you’d like to learn more about how Leader24 addresses the topics covered, here are some resources to get you started:

Article written with the help of AI.

Ready to transform your customer service?

Activate your AI assistant on WhatsApp in 5 minutes. 30-day free trial, no credit card required.

Share