A new phase of the European Union’s AI Act took effect Aug. 2, moving transparency obligations for chatbots, synthetic media and certain biometric systems from implementation planning into enforceable practice.
Providers must design covered systems to tell people when they are interacting with AI. They must also add machine-readable marks to covered AI-generated or manipulated content. Organizations deploying systems face additional disclosure duties for deepfakes, emotion recognition, biometric categorization and AI-generated material about matters of public interest when it lacks human review or editorial control.
The European Commission published guidance in July to clarify who is covered and how the rules should work. The requirements are part of a staged AI Act timeline rather than the arrival of the entire law at once, but they affect some of the most visible uses of generative technology.
Compliance becomes an interface
Transparency cannot live only in a legal memo. A chatbot needs a disclosure that a person can understand at the right moment. A generated image needs a durable technical signal and, in some cases, a visible label. A publisher or advertiser needs to know when human review changes the obligation.
That makes the work cross-functional. Product teams determine where notices appear. Engineers preserve metadata. Creative teams manage labels across edits and exports. Communications leaders decide how to describe automated production without confusing audiences. Legal teams must connect all of those decisions to the Act’s categories.
The machine-readable requirement is especially important because visible labels can be cropped or separated from content. Technical marking supports automated detection and platform-level handling, though no provenance system survives every transformation. Companies need a chain of custody that records how a piece of media was created, modified and distributed.
The rules also force a more precise conversation about “human in the loop.” A token approval step should not automatically be treated as meaningful editorial control. Organizations will need to document who reviewed the material, what they were responsible for and whether they had enough information to catch an error.
For companies outside Europe, the practical reach may be wider than the legal boundary. Global products rarely maintain entirely separate design and content systems for one market, and a clear EU standard can become the default elsewhere.
The AI Act’s transparency phase is therefore a product-design event as much as a regulatory one. The companies that handle it well will make disclosure legible without making every interaction feel like a warning label. The ones that wait for enforcement will discover that provenance cannot be added cleanly at the end.
About Julie: Julie Wohlberg is a journalist, communications strategist and entrepreneur with more than 25 years of experience building brands and launching companies. A former Tribune news and feature writer and early executive at Fotolia, she writes about media, brand, consumer behavior and the technologies reshaping how companies grow. She is the Founder + Principal of Inflection Point, a founder-led, senior growth strategy partnership for startups and growth stage companies.