Tallinn: As the first hard deadlines of the Artificial Intelligence Act approach, the Union is discovering that writing rules is easier than making them workable. From 2 August the transparency obligations in Article 50 begin to bite, requiring that people be told when they are dealing with a machine and that synthetic images, audio and text be marked as such. Turning that principle into something engineers can actually build has proved slower than the calendar allowed.
The Commission has leaned on codes of practice to bridge the gap. A voluntary code for providers of general-purpose AI models was finalised last summer, giving companies a way to show they meet their obligations before formal technical standards exist, which are not expected until 2027 at the earliest. A second code, this one covering the detection and labelling of artificially generated or manipulated content, moved through successive drafts in December and March before its final version was readied in late June. Neither code is legally binding. Both function as a passport of good faith, evidence a company can wave to regulators while the harder law is still settling.
The timing matters because the rulebook itself is shifting underneath everyone. Under the Digital Omnibus, a simplification package on which negotiators reached political agreement on 7 May, the deadline for the most sensitive high-risk systems listed in Annex III has been pushed back from August this year to December 2027. Supporters call it breathing room for firms drowning in overlapping obligations from the GDPR, the Data Act and the cybersecurity rules. Critics see a rulebook being loosened before it has been tested, with each delay chipping at the credibility of a law sold as the world’s strictest.
For a digitally minded state such as Estonia, the labelling question is not academic. Public services here run on data, and citizens who transact with government online will soon expect a clear signal when an answer, a summary or an image was produced by a model rather than a person. A watermark that is easy to strip, or a disclosure buried in fine print, would satisfy the letter of the law while defeating its purpose.
The deeper tension is between two things the Union wants at once. It wants to be the place that sets the global standard for trustworthy technology, and it wants its own companies to compete against American and Chinese rivals that face no such constraints. Every deferral and every voluntary code is an attempt to hold both goals together.
Whether the labelling regime earns public trust will depend less on the elegance of the legal text than on whether an ordinary user, scrolling through a feed, can tell in a glance what a machine made. On that unglamorous test the credibility of the entire framework now rests, and the answer is still being drafted.
