Enforcement of the European Union's Artificial Intelligence Act started on August 2. The European Commission's announcement says its AI Office, "together with national authorities, will begin enforcing" the law from that date, and that new transparency rules "start to apply" the same day [1].
Two obligations are named. Chatbots and other interactive AI systems "will have to tell users they are dealing with AI, not a human" [1]. Deepfakes, which the Commission glosses as "images, videos, or audio that have been edited or generated using AI," have to be labelled, and AI-generated or altered content has to "carry machine-readable marks so it can be detected more easily" [1].
Those are two different requirements doing two different jobs, and treating them as one is the easiest mistake to make about this law. A disclosure is aimed at a person. The support bot has to say it is a bot before the customer on the other end assumes otherwise, and any user can tell whether that happened. A machine-readable mark is aimed at software. It lets a platform, a newsroom verification desk or a court sort synthetic material at scale without anyone squinting at pixels, which is the only method that works once the volume gets past what humans can review. The first requirement is checkable by whoever is reading the screen. The second is checkable only by systems downstream that choose to read the mark, and the announcement does not say who is obliged to build that reading into their products.
The Commission frames the purpose as reducing "deception and manipulation" and helping people "make informed choices," while giving businesses "clearer obligations and a practical way to show compliance" [1].
The practical way is a code. The Commission has published a first list of more than 180 organisations that have signed the Code of Practice on transparency of AI-generated content, which the announcement presents as the instrument that operationalises the transparency rules [1]. What the announcement does not say is whether signing is voluntary, what a signature buys a company facing an enforcement action, or which of the 180 are the large model providers rather than downstream users of somebody else's model.
Here is what the announcement does not establish. It cites no article numbers from the AI Act. It names no penalty, fine or sanction of any kind. It sets out no later phase-in date, which means a reader cannot tell from this page which obligations are live today and which arrive in a future year. It does not distinguish between the provider that builds a system and the deployer that puts it in front of the public, and that distinction is what decides who is on the hook when an unlabelled deepfake circulates. It names no national authority, and it does not say how a person who encounters an undisclosed chatbot complains, or to whom.
Those answers live in the text of the AI Act and in the Commission's own guidance, which the announcement links: Guidelines on Transparency of AI-generated Content, the General-Purpose AI Code of Practice, guidelines for providers of general-purpose AI models, and guidelines on prohibited AI practices [1]. Reading the announcement settles the date, not the fine print.
The date is the part worth holding onto. As of August 2 the European position is that a person talking to a machine is entitled to be told, and that synthetic media should carry a mark whether or not a viewer can see it. The United States has no federal equivalent. The nearest American analogue this desk has covered moves in the opposite direction, criminalising one specific use of the technology rather than requiring the output to identify itself.
