Blog · Compliance · 30 July 2026 · 8 min read

The AI Act Deadline That Just Slipped Isn't the One That Protects You

AI-generated. This article was written by an AI system and published without substantive human review. Facts are drawn from primary sources and linked inline — verify anything you intend to act on.EU AI Act Art. 50(4)

On 27 July 2026 the Digital Omnibus on AI entered into force across the EU. The Commission's own announcement is unglamorous about it:"On 27 July 2026, the AI Omnibus enters into force across the EU, bringing extended timelines to administrative simplification."It is the largest amendment to the AI Act since the Act was adopted, and the coverage since has compressed into a single sentence: the EU blinked, the AI Act is delayed.

That sentence is true about one set of obligations and misleading about the set most people will actually meet. Three days from now, on 2 August 2026, a different part of the same law starts applying on its original schedule, untouched by the Omnibus. If you read this week's headlines and concluded the pressure is off, you have the wrong deadline in mind.

The obligations that just slipped are the ones that would have made your employer's AI prove itself. The obligation that did not slip is the one that tells you the AI is there at all.

whatsmyedge, July 2026

Three deadlines, routinely collapsed into one

Almost every error on this topic comes from treating the AI Act as having a single start date. It has several, and the Omnibus moved some and not others.

ObligationApplies fromChanged by the Omnibus?
Article 50 — transparency: chatbot disclosure, AI-content marking, deepfake labelling2 August 2026No. Unchanged.
Annex III high-risk — standalone systems (hiring, credit scoring, workplace monitoring)2 December 2027Yes. Moved back roughly 16 months.
Annex I high-risk — AI embedded in regulated products (machinery, toys, lifts)2 August 2028Yes. Moved back two years.

The Commission's guidelines page for the transparency obligations states the position in one line — "These transparency obligations apply from 2 August 2026" — with the guidelines themselves published on 20 July 2026 and the page touched again on 29 July. Nothing in the Omnibus moved that date.

The Omnibus did more than shift dates. It added a new prohibited practice under Article 5, banning AI systems that generate non-consensual sexually explicit imagery or child sexual abuse material. It broadened the permission to process special categories of personal data for the specific purpose of detecting and correcting bias. And it extended the AI Office's oversight to certain systems built on general-purpose models and embedded in large online platforms and search engines. A pure deregulation story this is not.

What the delay actually costs you

Annex III is the category that covers AI used to screen job applicants, rank candidates, inform promotion and termination decisions, and monitor workers. The obligations that were deferred to December 2027 are the ones that would have forced those systems to demonstrate they work: bias testing, technical documentation, event logging, and meaningful human oversight.

Nothing about those tools becomes illegal in the meantime, and nothing about them becomes safer. What changes is who carries the burden of noticing. For another sixteen months, if an AI system contributes to a decision about your role, there is no EU-wide duty on your employer to hold documentation showing how it reached that conclusion, or evidence that it was tested for bias.

The practical consequence for an individual is unglamorous and worth stating plainly: keep your own record. Where a decision about your work is informed by a tool, note what the tool was, what it produced, and what a human added on top. We haveargued before that exposure is a property of task structure rather than job title; the corollary is that the evidence of your judgment being load-bearing is something you can document, and for longer than expected, nobody else is required to document it for you.

What you actually get on Sunday

Article 50 is a disclosure regime, and its value to an individual is precisely that it removes ambiguity about what you are dealing with.

From 2 August, providers must ensure people are informed they are interacting with an AI system unless that is obvious from the circumstances. Synthetic audio, image and video audio, image, video or text content must be marked in a machine-readable format. Deepfakes must be disclosed as artificially generated. And AI-generated text published to inform the public on matters of public interest must be labelled, unless it underwent human review with someone holding editorial responsibility for it.

That last carve-out is the one most commonly misread in both directions. It does not mean every AI-assisted sentence on a company website needs a badge — ordinary commercial copy is not published to inform the public on a matter of public interest, and human editorial review removes the duty anyway. It also does not mean the rule is toothless: under Article 99 the fine ceiling for transparency breaches is EUR 15 million or 3% of worldwide annual turnover,whichever is higher whichever is higher for a large undertaking — but Article 99(6) flips this for SMEs and start-ups, capping them at whichever is lower, with Recital 165 requiring that a fine not threaten a small company's viability.

What to do with this

If you are an employee: assume the disclosure rule is now your main source of reliable signal about where AI sits in your workflow, and assume the high-risk safeguards are not arriving for another year and a half. Those two facts point the same direction — document your own contribution to decisions, because the regulatory backstop that would have generated that record is late.

If you run or advise an organisation: the compliance question due this week is small and concrete. Does anything you operate talk to people as if it were a person? Do you publish AI-generated public-interest content without human review? Do you produce synthetic media that could be mistaken for a real recording? If the answer to all three is no, Article 50 costs you nothing on Sunday. If the answer to any is yes, it is a labelling job, not a project.

And it is worth being precise about what happened this week, because a lot of commentary is not. No new law was adopted on 28, 29 or 30 July. The Omnibus became binding on the 27th. What has filled the days since is reaction — including from national governments now positioning themselves on enforcement. The delay is real, it is significant, and it applies to a category of system most people never knowingly touch. The rule that governs the AI you talk to every day did not move an inch.

Frequently asked questions

Was the EU AI Act delayed?

Part of it. The Digital Omnibus on AI entered into force on 27 July 2026 and moved the high-risk obligations back: standalone Annex III systems now apply from 2 December 2027 instead of 2 August 2026, and high-risk AI embedded in physical products under Annex I from 2 August 2028. The transparency obligations in Article 50 were not touched and still apply from 2 August 2026. "The AI Act was delayed" is true of one part and false of the part most people actually encounter.

What does Article 50 actually require?

Disclosure, not approval. People must be told when they are interacting with an AI system unless it is obvious; AI-generated or manipulated audio, image, video or text content must be marked in a machine-readable way; deepfakes must be disclosed; and AI-generated text published to inform the public on matters of public interest must be labelled unless it went through human editorial review. It is a labelling and honesty duty, not a licensing regime.

Does this affect the AI my employer uses for hiring or performance reviews?

It delays the rules that would have governed it. AI used for recruitment, promotion decisions and workplace monitoring falls under Annex III, the standalone high-risk category — the one that just moved from August 2026 to December 2027. The obligations that were postponed include bias testing, documentation, logging and human oversight requirements. The tools stay legal in the meantime; what is deferred is the duty to prove they work as claimed.

What are the penalties?

Under Article 99 of the AI Act, breaching the Article 50 transparency obligations carries administrative fines of up to EUR 15 million or 3% of total worldwide annual turnover for the preceding financial year, whichever is higher. Breaching the Article 5 prohibited practices carries up to EUR 35 million or 7%. Those are the figures for large undertakings. Article 99(6) inverts the calculation for SMEs and start-ups: for them each fine is capped at whichever of the two is lower, which for a small company means the turnover percentage rather than the headline millions. Recital 165 adds that fines must be proportionate and must not threaten the viability of an SME. These are ceilings applied with proportionality — not a schedule of expected fines.

Related reading

Where do you personally stand?

Take the free assessment

7 questions, scored in your browser, free.

Last updated 3 August 2026.