The European Union just switched on one of the most far reaching transparency rules in the world for AI content. From Sunday, August 2 2026, companies must visibly label AI generated images, audio, video and text that could pass as real, and stamp them with a hidden digital watermark. Chatbots must now tell users they are talking to a machine, unless that is already obvious.
This is not a small tweak. It is part of the EU AI Act, the broad law the bloc has been rolling out in stages since 2024. Existing AI tools already on the market get until December 2 2026 to add the watermarking pieces, but anything new launching now has no grace period at all.
The trigger for this rule was political. Fake audio of a Slovakian opposition leader appearing to plot election fraud, a mocked up video of France's president dancing in a nightclub, and a fabricated clip of the former UK Labour leader abusing staff all spread before anyone stepped in to stop them. Lawmakers built this rule to make that kind of content instantly recognizable as fake, protecting elections as much as consumers.
Here is the part most business owners are missing: this rule does not stay inside politics and social media. It reaches into advertising, marketing, publishing and film, anywhere AI is already used to make images, voiceovers or written copy that looks human made. The industry's own trade group, the Computer and Communications Industry Association, is warning that the definition of what counts as a labelable deepfake has been drawn so broadly that a stock photo touched up by AI in an advert could need the same label as a manipulated political speech.
The fines are real money. Violations can cost up to 15 million euros or 3 percent of a company's entire global revenue, not just its EU sales. And like the EU's privacy law GDPR before it, this rule follows the market, not the company's address. A business based in the United States, India or Singapore is still covered if its AI generated content reaches people in the EU.
Other governments are moving too, just differently. China already forced AI platforms to label generated content back in 2023 and tightened those rules further in 2025. California tried a similar law aimed at election deepfakes, and a federal judge struck it down over free speech concerns. The EU's approach sits in between: broader than the US will currently tolerate, but with less direct state control than China.
For any business using AI to generate marketing images, ad voiceovers, product descriptions or a customer facing chatbot, the sensible move is the same one companies made when GDPR arrived: audit every place AI touches your customer facing content now, before a regulator or a competitor points it out for you. The label itself is cheap. Finding out you needed one after the fact is not.