Chapter 7 of 8

High-risk deadlines: plan backwards from 2 December 2027

14 min read

The Digital Omnibus defers Annex III high-risk obligations to 2 December 2027 — agreed but not yet law. See what already binds and how to plan back from 2027.

Under the Digital Omnibus political agreement of 6–7 May 2026, the stand-alone high-risk obligations of Regulation (EU) 2024/1689, meaning systems classified under Article 6(2) and listed in Annex III, move from 2 August 2026 to 2 December 2027. Product-embedded high-risk systems under Article 6(1) and Annex I move from 2 August 2027 to 2 August 2028.

One caveat comes before anything else. As of June 2026 this is agreed but not yet law. The amendment still needs a European Parliament plenary vote, formal Council adoption and publication in the Official Journal. Until then, Article 113 of the Regulation as published still reads 2 August 2026. For the running legal status, see is the EU AI Act delayed.

The date moved; the workload did not. A 16-month deferral is schedule, not reprieve. This chapter shows how to use it: plan backwards from 2 December 2027.

The Corrected High-Risk Timeline at a Glance

The original Article 113 schedule made the high-risk regime applicable from 2 August 2026 for stand-alone Annex III systems and 2 August 2027 for Annex I product-embedded systems. The Digital Omnibus agreement moves those two dates, and only those two dates, to 2 December 2027 and 2 August 2028 respectively.

Three points govern everything that follows.

The new dates are fixed calendar dates. Negotiators rejected the standards-contingent 'stop the clock' proposal. The deferral does not depend on harmonised standards being available, and it does not move again if standards arrive late.
The deferral is agreed, not adopted. Until the amending regulation is published in the Official Journal, Regulation (EU) 2024/1689 as published is the law in force, and it reads 2 August 2026 for Annex III high-risk systems.
Nothing else moved. The prohibitions, the GPAI obligations and most transparency duties keep their original dates, and the Omnibus added a brand-new deadline on 2 December 2026.

Sixteen extra months sounds generous until you sequence the work: classification, conformity route, technical documentation, testing, oversight design, quality management, declaration, registration. Counted backwards from 2 December 2027, those months are already spoken for.

What the Digital Omnibus Agreed — and Why It Is Not Law Yet

Political Agreement, 6–7 May 2026

The Digital Omnibus is the Commission's digital simplification package of 19 November 2025, covering several files in the EU digital rulebook at once. Parliament and Council negotiators reached a provisional political agreement on 6–7 May 2026, and COREPER confirmed the compromise text around 13 May 2026.

For the AI Act, the agreement is a targeted amendment to the application schedule in Article 113, and nothing more. The substance of the high-risk obligations in Articles 8–17 is unchanged. The same risk management, the same documentation, the same conformity assessment, just later. For the full package beyond the AI Act dates, see the Digital Omnibus explained.

Fixed Dates: the Stop-the-Clock Was Rejected

The alternative on the negotiating table was a 'stop the clock': making the high-risk application date contingent on harmonised standards being available. Negotiators rejected it and chose certainty over flexibility. So do not plan as if the date might slip again when standards run late. 2 December 2027 is a calendar date, not a condition.

The Remaining Procedural Steps

Three steps separate agreement from binding law: a European Parliament plenary vote, formal adoption by the Council, and publication in the Official Journal. Until publication, the statute in force reads 2 August 2026 for Annex III high-risk systems.

Plan to the agreed dates, which are politically settled, but track adoption status. There is an asymmetry worth holding in mind: if the file unexpectedly stalled, the unamended 2 August 2026 date would still stand. That scenario is unlikely, but it is one more reason the deferral cannot justify a pause.

Before and After: the High-Risk Deadline Table

The table separates the two legal positions. The middle column is the law in force today; the right column is the agreed amendment pending adoption.

Provision	Statute as published (in force today)	Digital Omnibus agreement (pending)
Article 6(2) stand-alone high-risk — Annex III use cases: biometrics, employment, education, essential services, law enforcement and more	2 August 2026	2 December 2027 (+16 months)
Article 6(1) product-embedded high-risk — Annex I harmonised product legislation: machinery, medical devices, toys, lifts and more	2 August 2027	2 August 2028 (+12 months)
Article 50 machine-readable content marking and watermarking, plus the new CSAM/'nudifier' prohibition added by the Omnibus	2 August 2026 (marking duties)	2 December 2026

For every date in the Act in one place, including those already passed, see the full deadline timeline.

What Did Not Move — and Already Binds

Prohibitions: in Force Since 2 February 2025 — Article 5

The Article 5 prohibited practices have applied since 2 February 2025: social scoring, certain biometric categorisation, emotion recognition in workplaces and education, untargeted facial-image scraping, and the rest of the banned list. They carry the top penalty tier, up to EUR 35 million or 7% of total worldwide annual turnover under Article 99(3), and the Omnibus does not touch them.

GPAI Obligations: in Force Since 2 August 2025 — Articles 51–55

General-purpose AI model obligations under Articles 51–55 have applied since 2 August 2025 and are untouched by the Omnibus. If your company provides a GPAI model, the high-risk deferral gives you nothing.

Transparency and the New 2 December 2026 Date — Article 50

Most Article 50 transparency obligations still apply from 2 August 2026: disclosing chatbot interactions to users, and deepfake labelling duties for deployers. What moved is narrower. Machine-readable content-marking and watermarking duties land at the new 2 December 2026 date, alongside a newly added prohibition on CSAM and 'nudifier' applications. Many timelines published before May 2026 do not show that date at all.

The '2027' headline is therefore a trap. A provider of an Annex III recruitment system may simultaneously run a customer-facing chatbot (Article 50, August 2026) and generate synthetic content (December 2026). A compliance calendar that shows only one date will miss the two that arrive first. See how enforcement phases in for which authority enforces what, from when.

Planning Backwards from 2 December 2027

The method is subtraction. Start at 2 December 2027 and work upstream. Registration requires a signed declaration, the declaration requires a completed conformity assessment, the assessment requires Annex IV documentation, and the documentation requires evidence that takes quarters to produce, not weeks. Four steps, in dependency order; the full walkthrough lives at conformity assessment.

Step 1: Lock Your Classification — Article 6

Confirm whether each system is in Annex III scope under Article 6(2), and whether the Article 6(3) filter takes it out. A system that performs a narrow procedural task, improves a previously completed human activity or does only preparatory work may escape high-risk status if it poses no significant risk of harm, with the assessment documented in writing.

The Commission's draft classification guidelines under Article 6(5), published 19 May 2026 with a targeted consultation open until 23 June 2026, give practical in-and-out examples for both Annex I and Annex III. Use them as a sense-check, not as gospel. They are draft, not legally binding, the final version is expected later in 2026, and authoritative interpretation rests with the Court of Justice of the EU.

Step 2: Pick the Conformity Route — Article 43

Most Annex III providers self-assess under internal control (Annex VI). Remote biometric identification and certain other Annex III point 1 systems can require a notified body (Annex VII) under Article 43(1). Notified-body capacity is the scarcest resource on this timeline. If your route involves one, first contact belongs two to four quarters before the assessment, not after the documentation is finished. Article 43 conformity routes walks through the decision.

Step 3: Build the Annex IV Technical Documentation — Article 11

The Annex IV technical documentation, required by Article 11, is where the whole programme comes together. It must evidence the risk management system (Article 9), data governance over training, validation and testing data (Article 10), accuracy, robustness and cybersecurity testing (Article 15), automatic logging (Article 12) and human-oversight measures (Article 14). Each of those must be built and running before it can be documented, which is why this step cannot start in the final quarter.

Step 4: Testing, Logging and Human-Oversight Design — Articles 12, 14 and 15

Human oversight under Article 14 is a design obligation, not a policy document. Interface changes, override mechanisms and operator training materials need engineering sprints in the plan. The same goes for Article 12 logging, where the system must technically allow automatic recording of events over its lifetime, and the Article 15 test campaigns. Then come the wrap-up items that gate market placement: a quality management system (Article 17), the EU declaration of conformity (Article 47), CE marking (Article 48), and registration in the EU database for Annex III systems (Article 49).

A Quarter-by-Quarter Backward Plan

H2 2026: Classify, Scope, Assign Owners

AI inventory, Annex III mapping with the draft Article 6(5) guidelines as a sense-check, gap analysis against Articles 8–15, choice of conformity route, and a named owner for each workstream. For biometric systems, make first contact with a notified body now, not next year.

H1 2027: Build Documentation and Controls

Annex IV documentation buildout, risk management iterations under Article 9, test campaigns for accuracy and robustness under Article 15, human-oversight implementation under Article 14, and a stood-up quality management system under Article 17.

Q3 2027: Assess, Declare, Register

Run the conformity assessment, either Annex VI internal control or the Annex VII notified-body procedure, then sign the EU declaration of conformity (Article 47), affix CE marking (Article 48) and register in the EU database (Article 49).

Q4 2027 Is Margin, Not Working Time

The last quarter stays deliberately empty: buffer for notified-body queries, corrective findings, re-tests and late procurement evidence requests. If the buffer is consumed by planned work, the plan has already failed.

For a single Annex III system with an existing engineering team, expect four to six quarters of part-time-to-dedicated work. Multi-system portfolios scale roughly linearly on documentation and testing.

Set Your Internal Deadline at Mid-2027

The one operational takeaway is this: set an internal readiness target of mid-2027, roughly five months ahead of 2 December 2027, so audit findings, notified-body queues and re-tests land inside your margin instead of past the statutory date.

The commercial reality reinforces the legal one. Enterprise buyers and public procurement teams are already writing AI Act readiness into RFPs ahead of the statutory date, so 'ready by the deadline' often means 'too late for the deal'.

A mid-2027 target also hedges the legal-status uncertainty. Your plan no longer depends on exactly when the Omnibus amendment is published in the Official Journal, because you are ready well before either candidate date matters.

Keep one principle in view: the statutory deadline is the date enforcement can begin, not the date to aim for. Your own deadline comes first.

If You Miss It: Penalties and Enforcement

Non-compliance with the high-risk obligations, meaning Articles 8–17 and the operator duties in Articles 23–27, carries administrative fines up to EUR 15 million or 3% of total worldwide annual turnover, whichever is higher, under Article 99(4).

The other tiers set the contrast. Article 5 prohibited-practice breaches reach EUR 35 million or 7% (Article 99(3)), and supplying incorrect, incomplete or misleading information to notified bodies or national competent authorities reaches EUR 7.5 million or 1% (Article 99(5)). For SMEs and start-ups, Article 99(6) caps each tier at the lower of the percentage or the fixed amount.

Fines are not the only exposure. Market surveillance authorities can order corrective action, withdrawal or recall of non-conforming systems under Chapter IX, and for a shipped product, a recall is usually costlier than the fine that accompanies it.

The decision you need to make: pick your internal readiness date, with mid-2027 as the defensible default, put it in writing, and assign an owner to every workstream in the backward plan above. The deferral bought you sixteen months of schedule. Converting that schedule into compliance is your decision, and it is due now, not in 2027.

How Confir helps

The backward plan above is what Confir operationalises. Confir's classification module asks plain-English questions about what each AI system does and derives the risk tier under Articles 5 and 6, including the Article 6(3) filter assessment, documented in writing. From the classification result, the platform maps which obligations apply, generates the Article 11 / Annex IV technical documentation pack, assembles the conformity-assessment evidence trail, and produces the Article 47 declaration of conformity.

The synthesis engine is deterministic and rule-based: the same logic every time, no model inference, no hallucination. Every classification decision and its rationale lands in an immutable audit log, the evidence a market surveillance authority or a notified body will ask for first.

Frequently Asked Questions

Is the EU AI Act high-risk deadline delayed to 2027?

Politically, yes. The Digital Omnibus agreement of 6–7 May 2026 moves stand-alone Annex III high-risk obligations from 2 August 2026 to 2 December 2027. But it is not yet law: it still needs a European Parliament plenary vote, Council adoption and Official Journal publication. Until then, Regulation (EU) 2024/1689 still reads 2 August 2026.

When do high-risk AI systems have to comply with the EU AI Act?

Once the Digital Omnibus becomes law, stand-alone Annex III systems (Article 6(2)) must comply by 2 December 2027 and product-embedded Annex I systems (Article 6(1)) by 2 August 2028 — both fixed calendar dates. As of June 2026 the statute as published still says 2 August 2026 and 2 August 2027, because the amendment has not yet been published.

What is the Digital Omnibus and what does it change in the AI Act?

The Digital Omnibus is an EU amending package that reached provisional political agreement on 6–7 May 2026, with the text confirmed in COREPER around 13 May. For the AI Act it defers the two high-risk application dates to 2 December 2027 and 2 August 2028, adds a 2 December 2026 content-marking and CSAM date, and rejects the standards-contingent stop-the-clock.

Is the whole EU AI Act delayed?

No. The Article 5 prohibitions have applied since 2 February 2025, GPAI model obligations under Articles 51–55 since 2 August 2025, and most Article 50 transparency duties still apply from 2 August 2026. A new 2 December 2026 deadline covers content marking and the CSAM ban. Only the two high-risk application dates move.

What are the penalties for missing the EU AI Act high-risk deadline?

Non-compliance with high-risk obligations carries fines up to EUR 15 million or 3% of worldwide annual turnover under Article 99(4). Prohibited practices draw up to EUR 35 million or 7% (Article 99(3)), and supplying incorrect or misleading information to authorities up to EUR 7.5 million or 1% (Article 99(5)). For SMEs and start-ups, Article 99(6) caps fines at the lower amount.

How long does conformity assessment take for high-risk AI?

Plan in quarters, not weeks. Most Annex III providers self-assess under internal control (Annex VI), but the groundwork — Annex IV documentation, testing evidence, human-oversight design and a quality management system — typically takes two to four quarters. Biometric systems under Annex III point 1 may need a notified body, whose queues add months. Target internal readiness by mid-2027.

Was the EU AI Act stop-the-clock proposal adopted?

No. The proposal to tie high-risk application dates to the availability of harmonised standards was rejected in the Digital Omnibus negotiations. The agreed deferrals — 2 December 2027 for Annex III systems and 2 August 2028 for Annex I product-embedded systems — are fixed calendar dates that do not move if standards arrive late.