On August 2, 2026, the transparency rules in Article 50 of the EU AI Act began applying, and EU authorities gained enforcement powers the same day. Most coverage is written for lawyers or for large EU companies. This is the plain English version for a US business owner: what the rules actually require, when the remaining deadlines hit, what might reach you across the Atlantic, and a five question check to see where you stand. One note before anything else: this is educational content, not legal advice. If the check below raises flags for your business, that is your cue to talk to counsel, not to rely on a blog post.
Article 50 creates disclosure duties in three main buckets. First, AI systems that interact directly with people, like chatbots, must be designed so users know they are dealing with AI. Second, providers of AI that generates or manipulates text, images, audio, or video must make that content machine detectable through markers or watermarking. Third, deployers must disclose deepfakes and AI generated text published to inform the public on matters of public interest, unless a human takes genuine editorial responsibility for it. The European Commission’s guidance is explicit that the human review has to be substantive. Assistive uses, like standard editing that does not materially change content, sit outside the strictest duties.
It can, in some circumstances, and the honest answer is that the edges are untested. The Act applies to providers and deployers whose AI systems or outputs are placed on the EU market or used in the EU. Legal analyses note that deployer duties can apply where outputs are foreseeably disseminated in the EU, and content posted on the globally accessible internet can qualify. Penalties for transparency violations can run up to 15 million euros or 3 percent of global annual turnover, whichever is higher.
What nobody knows yet is how enforcement will treat small non EU businesses in practice. Regulators just received these powers, and there is no track record. The sensible posture is not panic and not dismissal: if you serve EU customers, run ads reaching the EU, or publish AI generated content to the open internet, you should know where you stand rather than assume the answer.
Three dates. August 2, 2026: Article 50 transparency obligations apply and enforcement powers are active. December 2, 2026: marking and detection requirements extend to generative AI systems that were already on the market before August 2. February 2, 2027: a watermark interoperability commitment applies for signatories of the EU’s voluntary Code of Practice. One more wrinkle worth knowing: individual EU member states can layer additional national transparency rules on top of the Act, so the baseline is not the ceiling.
Question 1: Do you use a customer facing chatbot, voice agent, or AI email responder, and does it clearly identify itself as AI? A yes and yes means you match the rule’s core expectation. A yes and no is the most concrete gap on this list.
Question 2: Do you publish AI generated informational content, and would a reader know? If a human genuinely reviews, edits, and takes responsibility for it, the strictest labeling duty does not apply. If content goes out essentially as the machine wrote it, it is label territory.
Question 3: Do you publish AI generated or AI altered images, audio, or video of real people, places, or events? Realistic synthetic media is the deepfake category, and it carries disclosure duties with narrow exceptions.
Question 4: Do you serve EU customers, ship to the EU, run ads that reach the EU, or publish to audiences that include EU users? This question does not create obligations by itself, but it determines how seriously to take your answers to the first three.
Question 5: Is there one named human who reviews and owns each piece of published AI assisted content? This is the workflow that both satisfies the exemption’s spirit and, separately, is just how trustworthy content gets made.
Your no answers are your to do list, roughly in the order above.
The EU published a voluntary Code of Practice recommending consistent labels such as “AI”, “AI Generated”, and “AI Modified”, and the Commission has confirmed that following the code is one accepted way to show conformity. On the technical side, content credential standards like C2PA are being adopted to carry machine readable provenance. Expect these labels to become familiar interface furniture over the next year, the way cookie notices did, and note that Meta has already been labeling AI content on its platforms independent of this law.
Several real open questions. How broadly “matters of public interest” will be interpreted for ordinary marketing content is not settled. What evidence of human editorial responsibility will satisfy a regulator is untested. Enforcement priorities, especially toward small businesses outside the EU, are unknown. And national add on rules are still emerging. Anyone giving you a definitive compliance verdict today is ahead of the facts. The durable move is the one that is right regardless of interpretation: honest labeling where content is synthetic, and a real human owning what you publish.
Questions, answered
No. Assistive use, like drafting help and standard editing, is treated differently from fully synthetic content, and text with genuine human editorial responsibility sits outside the strictest labeling duty. The review has to be substantive, not a glance.
Up to 15 million euros or 3 percent of global annual turnover, whichever is higher. How fines will be applied in practice, especially outside the EU, has no track record yet.
It can. The Act covers systems and outputs placed on the EU market or used in the EU, and legal analyses note that content foreseeably disseminated in the EU, including on the public internet, can bring deployer duties into scope. The practical reach for small non EU businesses is still untested.
A provider builds an AI system or has one built and puts it on the market under its own name. A deployer uses an AI system in the course of business. A small business using ChatGPT to make content is typically a deployer.
Whether you have a clear roadmap or just a sense that your digital potential is not being fulfilled yet, we will figure out what is possible, build what is needed, and push until the needle moves.