Article 50 Applied on Sunday. Here Is What Is Now Enforceable.

The date that generated a year of countdowns arrived on a Sunday, which is a fitting end to it. Nothing announced itself on 2 August 2026. No inbox filled up, no portal opened, no regulator appeared. Monday was the first working day under the transparency obligations in Article 50, and for most firms it looked identical to the Friday before, which is exactly the condition under which a change gets quietly ignored.

So it is worth being precise about what did and did not happen, because the surrounding commentary spent so long on the deferral debate that the part which actually landed has been under-reported. The high-risk regime moved to December 2027 and August 2028 under the Digital Omnibus. The transparency duties did not move, and they now apply.

What is now in force

Two obligations matter for professional teams. Where an AI system interacts directly with a person, that person has to be informed they are dealing with AI unless it is already obvious from context. And AI-assisted or manipulated content carries disclosure duties, with the sharpest attention on synthetic media and on text published to inform the public about matters of public interest, unless a human editor has taken editorial responsibility for it.

That last clause repays attention, because it points at something broader than a labeling rule. The obligation eases where a human has genuinely taken responsibility for the output, which is a drafting choice that says something about the whole direction of the regulation. Accountability substitutes for disclosure. If a named person stands behind the work, the requirement to flag it as machine-produced softens, because the thing the rule is actually protecting is the reader’s ability to know whether anyone is answerable.

The enforcement question is not the interesting one

The Act is backed by an enforcement regime, and the immediate instinct is to ask how aggressively it will be applied in the first months. That is a reasonable question and an unanswerable one, since it depends on national authorities that are still being stood up, and predicting regulator behavior is not something we are going to do here.

The more useful question is commercial rather than legal, because it does not wait for anybody to be designated. A date like this changes what counts as a normal question to ask a supplier. Procurement templates absorb the vocabulary, insurers update questionnaires, and a client asking how your AI-assisted work was reviewed stops sounding suspicious and starts sounding standard. That shift needs no enforcement action at all to reach you.

Two failure modes to avoid this month

The first is the rushed disclaimer: a line added to every deliverable stating that AI may have been used, applied indiscriminately and substantiated by nothing. It reads as anxiety rather than process, it tells the client nothing they can use, and if a dispute ever arrives it is a written claim about your workflow with no record behind it.

The second is the opposite, and more common: concluding that because your firm does not operate a high-risk system, none of this applies, and filing the whole subject under December 2027. Transparency duties and high-risk classification are separate questions, and the deferral of one says nothing about the other. Teams that conflate them are reading the calendar rather than the text.

The week two move

The productive thing to do now is narrow. Take the workflow where AI contributes most to what clients actually receive, and establish two facts about it: what evidence the output was grounded in, and who reviewed it before it went out. If both answers exist and are recorded somewhere durable, disclosure becomes a description of something real. If either answer is missing, that gap is the work, and it is more valuable than any wording exercise.

Qonera is the AI governance platform for professional teams, built around a structured review and approval workflow where evidence is audited before analysis, answers carry per-claim citations back to source, and a named reviewer approves the result before delivery, with each step written into a tamper evident audit trail. The article-by-article mapping is published on the EU AI Act page. August 2 has now passed without incident for almost everybody, which is the least interesting fact about it: the questions it made routine will arrive gradually, from clients rather than regulators, and they will be easier to answer for the firms that treated the date as a starting line.

This article is for general information only and does not provide legal advice. Organisations should consult qualified legal counsel about how Article 50 and the EU AI Act apply to their specific systems, workflows, and obligations.