Open Editor
Privacy
7 min read

The EU AI Act's Next Real Deadline Is December 2, 2026 — Not the One That Got Delayed

The headline that moved was a 16-month reprieve on high-risk AI rules. The deadline that didn't move — and lands in under three months — applies to any generative AI system already on the EU market.

DateWhat Happens
August 2, 2026Article 50 transparency obligations enforceable (already passed)
December 2, 2026Grace period ends: pre-existing generative AI systems must mark outputs
February 2, 2027Watermark-detection interoperability required for Code signatories
December 2, 2027Standalone high-risk (Annex III) systems must comply — delayed from Aug 2026

Anyone who's followed EU AI Act coverage in 2026 has likely seen the big story: the compliance deadline for high-risk AI systems got pushed back sixteen months, from August 2026 to December 2027. That's real, and it's the deadline most compliance planning has centered on. But it's not the next deadline. A narrower, less-discussed requirement — one the delay explicitly left untouched — arrives December 2, 2026, less than three months from now, and it applies to a much wider set of companies than the high-risk category ever did.

The Deadline That Actually Moved

The EU's Digital Omnibus on AI, first proposed by the European Commission in November 2025, reached political agreement on May 7, 2026, and received final Council approval on June 29, 2026. Its central change: standalone high-risk AI systems under Annex III — covering things like hiring tools, credit scoring, and systems used in education, law enforcement, or biometric identification — now have until December 2, 2027 to comply, instead of the original August 2, 2026 date. AI embedded in already-regulated products, like medical devices or machinery, got its own parallel extension, moving from August 2027 to August 2, 2028. Systems already on the market before their respective deadlines won't need to comply at all unless they're substantially modified afterward.

What the Delay Explicitly Did Not Touch

Here's the detail that gets lost in "the AI Act got delayed" headlines: the Digital Omnibus left Article 50 — the transparency and content-labeling rules — almost entirely alone. Those obligations, covering disclosure when someone is interacting with an AI system and labeling requirements for AI-generated or manipulated content, took effect on their original date of August 2, 2026, which has now passed. The only piece of Article 50 that got any grace period is the narrower marking-and-detection requirement in Article 50(2), and even that only shifted by four months — not sixteen.

It's easy to read "the AI Act's deadline moved to 2027" and assume the whole law slowed down. What actually happened is that one specific, heavily-covered category — high-risk systems — got real breathing room, while the transparency rules governing AI-generated content kept moving on their own, much tighter schedule.

The Real Next Deadline: December 2, 2026

Under the Digital Omnibus's grandfathering rule, generative AI systems that were already on the market before August 2, 2026 have until December 2, 2026 to bring their outputs into compliance with Article 50(2): marking synthetic audio, image, video, or text so it's detectable as AI-generated in a machine-readable format. This applies far more broadly than the high-risk category — any provider of a generative AI system already operating in the EU falls under it, not just companies building hiring or credit-scoring tools.

The technical detail worth knowing: the EU's Code of Practice on transparency, finalized under Article 50, requires a layered approach — generally both digitally signed metadata and an imperceptible watermark, rather than either one alone, since the Commission and AI Board concluded no single current technique meets all four of the law's quality requirements at once. That's the same conclusion covered in more depth in the earlier post on C2PA's adoption status, which goes into why cryptographic provenance and statistical watermarking are being treated as complementary rather than either one being sufficient alone. For free-form AI-generated text specifically, the Code sets a 200-token threshold — watermarking is required above it, with an exemption for very short text below that length.

The Deadline After That: February 2, 2027

A second, quieter date follows close behind. Providers who sign the EU's Code of Practice — the mechanism most companies are using to demonstrate Article 50 compliance — must have an interoperability solution for watermark detection in place by February 2, 2027. The point of this requirement is specific: a watermark only serves a transparency purpose if something other than the original provider's own tools can detect it. Without an interoperable detection path, content marked by one provider's system can't be verified by anyone using a different provider's detector, which defeats the purpose of a labeling requirement in the first place.

RequirementDeadlineApplies To
Mark outputs (new systems)Aug 2, 2026 (passed)Any generative AI system launched after this date
Mark outputs (existing systems)Dec 2, 2026Generative AI systems already on the market before Aug 2, 2026
Watermark-detection interoperabilityFeb 2, 2027Code of Practice signatories relying on watermarking

What Happens If a Provider Misses the Date

Non-compliance with Article 50 carries penalties of up to €15 million or 3% of global annual turnover, whichever is higher — a separate and lower tier than the up-to-€35-million or 7% penalties that apply to the most serious AI Act violations elsewhere in the regulation. Regulators, law enforcement, media organizations, fact-checkers, researchers, and civil society groups are entitled to free, unrestricted access to detection tools regardless of how the interoperability requirement is ultimately implemented — a provider can charge for high-volume commercial use of its detection tools, but not for those categories of users.

Why the Confusion Is Understandable

Three EU AI Act deadlines are converging in the same rough window — December 2026, February 2027, and December 2027 — and they cover genuinely different obligations aimed at different companies. It's a reasonable mistake to see "deadline pushed to 2027" reporting and conclude the whole law relaxed, when in practice the transparency and watermarking obligations that touch by far the largest number of AI providers are still on their original, much tighter track. For any company already generating or distributing AI content in the EU, the operative date to plan around right now isn't 2027 at all — it's December 2, 2026, three months out.


The honest summary: yes, a major EU AI Act deadline was delayed by sixteen months — but it was the high-risk systems deadline, which affects a comparatively narrow set of use cases. The transparency and watermarking rules under Article 50, which apply to nearly any provider of generative AI operating in the EU, are still moving on their original, much tighter calendar, with the next real deadline landing December 2, 2026 and a second one following on February 2, 2027.

For questions or inquiries contact us at info@cleartexteditor.com