The guidance arrived after the obligation
Fifteen days is not a scandal. But the sequence — duty first, instructions second — is the shape of AI regulation everywhere right now, and it explains why compliance keeps being performed rather than achieved.
The Commission published its transparency guidance on 17 August. The obligation it explains commenced on 2 August. Nobody was ambushed — the duty was legible from the statute long before either date — but for fifteen days the operative question was not "what does the law require" but "what will the regulator accept", and those are different questions with different answers.
This is not a European quirk
It is the standard sequence for regulating a moving target. The legislature writes a principle because a specification would be obsolete before enactment. The regulator then supplies the specification, later, when it can see what the industry has built. Every jurisdiction attempting AI rules is doing some version of this.
The cost is predictable: a window in which the diligent over-comply, the aggressive wait, and neither is wrong. Anyone who built to a strict reading of the text spent money they may not have needed to spend. Anyone who waited was right and looks reckless.
Which is why the enforcement that lands first is old law
The most instructive enforcement action this month came from a different direction entirely. The US Justice Department settled with OpenAI OpCo and Statsig over citizenship-status discrimination in recruitment workflows — under statutes that have existed for decades, with enforcers who have been doing this for decades.
No guidance gap there. No commencement date to argue about. The rule was already clear, the enforcer already knew how to apply it, and the only novelty was that a model sat inside the process.
The practical conclusion
An organisation watching Brussels for the rules that will bind it is watching the slowest-moving source. The fastest-moving one is the body of employment, credit, housing and consumer law that already covers automated decisions and already has people whose job is to bring cases. Those bodies do not need new powers, do not publish commencement dates, and are not waiting for guidance.
The EU work still matters — €15 million or 3% of worldwide turnover is not a rounding error, and the December 2027 high-risk regime is a genuinely large programme. But it is the risk you can see coming. The other kind is the one that arrives as a letter about something you did last year.
European Commission — Guidelines on transparency obligations under the AI Act → · Cooley — EU AI Act: Transparency Obligations Take Effect 2 August 2026 → · Holland & Knight — US Companies Face EU AI Act's August 2026 Compliance Deadline → · AI Governance Weekly — AI Governance Weekly — August 20, 2026 →