EU AI Labels Don't Start Sunday. They Started in August.
EU AI Labels Don't Start Sunday. They Started in August.
Wrong. Europe's AI-labelling rules do not "take effect on Sunday" — they took effect two months ago. A British tech site told its readers this week that tech companies "must label AI content from Sunday," presenting the EU's labelling, watermarking and metadata duties for AI-generated images, audio, video and text as a new deadline landing on October 4. Flat wrong. And we can trace exactly where the error came from, because the fingerprints of a two-month-old news-wire story are all over it. What is real is a different date entirely: December 2, when a narrow grace period closes and a fresh set of prohibitions kicks in. The distinction matters. Recycling stale deadlines is quietly becoming one of AI journalism's shoddiest habits.
The claim, in computing.co.uk's words
Published October 2, the computing.co.uk piece said EU rules on labelling AI-generated and AI-altered content "take effect on Sunday" — applying to firms operating in or targeting the EU, with carve-outs for artistic and satirical works. It added that "after [2nd December] regulators will be able to impose financial penalties." If you run a product team at a US tech company, that reads like an urgent compliance deadline 48 hours away. It reads that way because it was written to read that way. Unfortunately for everyone who planned their weekend around it, almost nothing in the framing survives contact with the actual law.
Now the law, as written
Here's the thing: the transparency obligations are Article 50 of Regulation (EU) 2024/1689 — the EU AI Act. And Article 113 of that regulation leaves no room for doubt: the Act applies from 2 August 2026, and Article 50 falls squarely under the general application date. The European Commission's own FAQ page on the Article 50 transparency obligations spells out the duties — providers must mark AI-generated content in machine-readable form, deployers must disclose deepfakes and AI-generated text — with none of the "coming Sunday" framing. We pulled the computing.co.uk piece against the Commission's FAQ and the consolidated legal text on eur-lex, and there is no reading of the regulation under which an October 4 deadline exists. None.
Could some intervening legislation have moved the date? The obvious candidate is the Digital Omnibus — Regulation (EU) 2026/1744, in force since July 27, 2026 — which did defer certain high-risk AI obligations. But it did not touch Article 50. The labelling duties have been live since August, full stop. When the Article 50 guidelines landed on July 20, the Commission's Henna Virkkunen — Executive Vice-President for Tech Sovereignty — said this: "The Commission supports the smooth and effective application of the AI Act to make AI systems interacting with people such as chatbots and AI agents and AI content more transparent and trustworthy… These guidelines support providers and deployers in meeting their obligations under the AI Act, while helping citizens know when they are interacting with AI." The tense gives it away: obligations. Already existing. Already applied.
The real origin of "Sunday"
That leaves the "Sunday" question. It came from July: on the 29th, AFP's Frédéric Pouchot filed a wire story — "Made by AI? EU tells firms to stick a label on it from Sunday," more or less. There, "Sunday" meant 2 August 2026. The genuine deadline. The computing.co.uk piece published this week carries the same anecdotes (the 2024 Taylor Swift deepfakes, the Arup £20 million deepfake scam, the 2022 Zelensky surrender video), the same December 2 date, the same artistic and satirical carve-outs, the same Karen Massin quote warning that too many labels could confuse users, and the same line about a "technical study combining multiple methods." It is the July wire, reheated, with the calendar never updated — and "Sunday" left dangling like a live wire.
This is not a subtle error. It is what happens when deadline journalism meets the content treadmill: a two-month-old story gets a fresh timestamp, and suddenly thousands of readers believe a law that has been in force since August is arriving this weekend. Sergey Lagodinsky, the Green MEP who helped negotiate the AI Act, told The Guardian back in July why the labelling regime matters: "It is a matter not only of customer protection, it's also a matter of democracy protection… Making transparent this information is something which we need to preserve our democracy and the authenticity of facts online." He deserves better than to have his cause misreported by calendar error.
December 2: real date, wrong story
Here is the part computing.co.uk got half-right, which makes the error more corrosive: 2 December 2026 is a real date, but it is not when "regulators will be able to impose financial penalties." Penalties have been enforceable since August 2 — up to €15 million or 3% of worldwide turnover for an Article 50 breach, enforced by national market-surveillance authorities with the Commission's AI Office in the loop. What December 2 actually marks is the end of a narrow grace period: generative AI systems placed on the market before 2 August 2026 get extra time to comply with the Article 50(2) machine-readable-marking duty (new Article 111(4), added via the Omnibus). Deployer deepfake-labelling duties had no grace period at all. December 2 is also when new prohibitions on non-consensual intimate imagery and CSAM-generation start applying. That is the genuine enforcement story — and it deserved a real article instead of a recycled one.
So what actually happened this week
We searched for anything genuinely new on AI-content labelling this week — fresh Commission guidance, a fine, an enforcement action, a company compliance announcement — and found nothing in the EU. The last real moves were the Article 50 guidelines (July 20), the voluntary Code of Practice on Transparency of AI-Generated Content (finalised June 10, with roughly 180–190 signatories by the end of July), and enforcement beginning August 2. Markus Reinisch, Meta's vice-president for public policy in Europe, captured the industry's grumble when signing the code in July: "We must avoid a situation where online content is subject to a growing array of different labels and disclosures that end up overwhelming people, while at the same time adding additional regulatory complexity for providers." Google's Karen Massin made the same point more bluntly, warning that "an increasing number of labels and disclosures could ultimately confuse users rather than improve transparency if they become too commonplace."
Compare that with the United States and the contrast is stark. On September 30, President Trump and six tech CEOs — Pichai, Amodei, Zuckerberg, Brockman, Musk and Huang — signed the White House Accord on Super Intelligence. Voluntary. "Morally binding." Four layers of internal controls and independent audits — no penalties, and no disclosure duties. Meanwhile 73% of Americans say they don't trust AI companies — a number we dug into this week — and Washington's answer is a handshake. Elsewhere, India moved: advertising regulator ASCI issued guidelines on September 29 mandating disclosure of synthetically generated content in ads, effective three months from publication, while the Supreme Court's stay on the Gujarat deepfake case comes back up on October 5 — a date actually worth watching next week, unlike October 4.
Why one bad headline matters more than you'd think
And here is our position: stale-deadline journalism is not a victimless error. Every time a reputable-looking tech site invents urgency around a law that has been in force for two months, three things happen. First, compliance teams waste cycles on a phantom deadline. Second, readers learn — correctly, this time — that tech journalism can't be trusted on regulatory facts, which is poison for the moment when a real deadline arrives. And third, the actual story — that December 2 closes the grace period for pre-existing systems and switches on new prohibitions — gets buried under the noise of the fake one. The EU's labelling regime is genuinely consequential: it is the world's first binding attempt to make synthetic content traceable at scale, it carries fines that can reach 3% of global turnover, and its interaction with America's voluntary approach will define the next decade of AI governance. That story deserves reporting, not reheating. The next time a headline tells you a European deadline lands "on Sunday," check the calendar — and then check eur-lex. We did. It took ten minutes.
EU AI labelling: the questions everyone asks
When did the EU's AI-labelling rules actually take effect?
2 August 2026, under Article 50 of Regulation (EU) 2024/1689 — the Article 113 general application date.
What changes on 2 December 2026?
The grace period closes for generative systems placed on the market before 2 August 2026 — the machine-readable marking duty applies (Article 111(4), via the Omnibus), and new prohibitions on NCII and CSAM generation start applying.
What are the fines?
Up to €15 million or 3% of worldwide turnover (the Article 99/50 tier) — penalties enforceable since 2 August 2026.
Do US companies have to comply?
Yes, if they operate in or target the EU. Article 50 has no US exemption.
Sources
computing.co.uk, "Tech companies must label AI content from Sunday" (October 2, 2026); AFP wire by Frédéric Pouchot (July 29, 2026); European Commission FAQ on transparency obligations under Article 50 of the AI Act (digital-strategy.ec.europa.eu); consolidated text of Regulation (EU) 2024/1689, Articles 50 and 113 (eur-lex.europa.eu); Regulation (EU) 2026/1744, the Digital Omnibus (in force July 27, 2026); Commission statement on Article 50 guidelines (July 20, 2026); The Guardian (July 2026); White House Accord on Super Intelligence (September 30, 2026); ASCI guidelines via storyboard18/business-standard (September 29, 2026).
Read next: Google pays publishers for AI search — most get under 0.1%.
Daniel Rao is the founder and editor of TechPulse USA, covering AI, chips and the geopolitics of the intelligence age — from data-centre financing to export controls.
Comments
Post a Comment