As of today, every chatbot on your website has to admit it isn’t human.
Since 2 August 2026, the transparency rules of the European AI Act are in force. Companies, public bodies and freelancers who use AI toward customers now have to make that clear, De Tijd reports. Chatbots, AI-generated images, videos and text all fall under a new labeling duty. Ignore it, and you risk a fine up to 15 million euro or 3 percent of global turnover.
Five obligations under Article 50
The rules sit in Article 50 of the AI Act and touch more businesses than you’d expect. A chatbot or AI helpdesk has to tell the customer from the first message that they’re talking to a machine. Images, audio or video made with generative AI that could pass for real need a visible or audible mark as artificial, separate from the technical, machine-readable watermark the AI provider itself has to add. Systems that detect emotions or categorize people biometrically have to inform the person affected. Deepfakes, realistic content that doesn’t match reality, need to be labeled as such, even without any intent to deceive. And AI-written text on matters of public interest gets a label too, unless a human carries out editorial review and takes responsibility.
There are exceptions for minor technical edits, clearly unrealistic content, older content and satire. That last one is bound to spark debate: what counts as satire and what doesn’t?
Plenty of companies know the AI Act exists, but underestimate how much AI is already running inside their own organization.
That’s according to lawyers Henri De Jonghe and Sibrecht Kinget of Wauters & Partners, speaking to De Tijd. They compare this phase to the rollout of GDPR: companies know the law exists, but not their own “shadow AI.” Outsourcing AI work to a marketing agency or customer service provider doesn’t automatically get you off the hook either. What matters is who actually decides how the AI content gets used.
Belgium still has no regulator
The European Commission published official guidelines on 8 May 2026 to clarify exactly who has to label what. Even so, enforcement is lagging in places. Belgium still has no legally designated AI regulator. BIPT, the telecom watchdog, is expected to take on that role, but formal parliamentary designation hasn’t happened yet. Companies that need to get their AI systems certified still don’t know where to go.
Worth knowing: the heaviest obligations, for so-called high-risk AI such as hiring and selection, are postponed to 2027 and 2028. That delay doesn’t apply to today’s transparency rules. The recent “Digital Omnibus,” which loosens parts of the AI Act, leaves Article 50 untouched too.
What does this mean for your marketing?
For a marketing team working with AI, this goes further than a chatbot. Think product images edited with AI, HR training videos, or influencer content. Once an image is photorealistic and indistinguishable from a real photo to the naked eye, it falls under the labeling duty, even if it’s an ad for a real product. We track this kind of development and share updates on LinkedIn, much like our piece on authenticity versus AI, where transparency already turned out to be the core issue.
Conclusion
Waiting for the regulator to show up isn’t a strategy. Map out where AI already sits in your organization now, and adjust your content and communication accordingly. Not sure where you stand, see what we can do for you.
Do you show up in ChatGPT and Perplexity?
Transparency runs both ways. While you’re labeling your own AI use, you also want to show up when customers ask AI tools about your industry. Test it for free with the AI visibility scan: we check whether your brand shows up in ChatGPT, Perplexity, Claude and Gemini, and deliver a report with concrete action points.