• Report
  • 7 min read

The AI Act’s December 2027 deadline: what data teams need in place

From 2 December 2027, AI used in hiring, credit scoring, insurance pricing and other Annex III areas must meet the AI Act’s high-risk rules. What the law asks of data teams, and what to build in the next twelve months.

ResearchBy HarshPublished 9 November 20267 min read5 sources

Key findings

  1. 1Obligations for high-risk AI in the areas listed in Annex III, such as recruitment, credit scoring and the pricing of life and health insurance, apply from 2 December 2027, after the Digital Omnibus moved the date from August 202612
  2. 2Much of the work sits with data: training, validation and test data must be relevant, sufficiently representative and, to the best extent possible, free of errors and complete, under documented governance practices3
  3. 3Organisations that only use a high-risk system still have data duties: input data they control must be relevant and representative, and the system’s logs must be kept for at least six months3
  4. 4The first European standard written for the Act, EN 18286 on quality management, was published in July 2026, and the Commission consulted on draft guidelines for classifying high-risk systems in May and June 202645

The date moved, the work did not

The Digital Omnibus on AI, published as Regulation (EU) 2026/1744 in July 2026, moved the obligations for stand-alone high-risk systems in Annex III from 2 August 2026 to 2 December 2027, and those for AI built into regulated products to 2 August 202812. The transparency rules in Article 50, including telling people when they are dealing with a chatbot, kept their August 2026 date1.

For a data team planning 2027, that leaves about a year. Most of what the high-risk rules require cannot be added at the end: it depends on records of where data came from, how it was prepared and how the system behaves in use, and those records are easiest to keep from the start. The dates for every EU rule that affects data and AI work are in our EU rules calendar.

Is your system in scope?

Annex III areas most likely to reach a data team
AreaExamples
EmploymentScreening or ranking candidates, decisions on promotion or termination, monitoring performance
Essential private servicesCreditworthiness assessment and credit scoring, risk assessment and pricing for life and health insurance
Public servicesDecisions on eligibility for public benefits and services
EducationAdmissions, assessing learning outcomes, monitoring tests
Critical infrastructureSafety components in digital infrastructure, road traffic and the supply of water, gas, heating and electricity
Biometrics, law enforcement, migration and justiceIdentification and categorisation of people, risk assessments, help with judicial decisions

Source: AI Act, Annex III. The annex lists eight areas; this table groups the last four.

A system in one of these areas is not high-risk if it does not pose a significant risk of harm, for example because it only performs a narrow procedural task or improves the result of work a person has already done. That exemption never applies to systems that profile people3. The Commission published draft guidelines on classification for consultation in May 2026, and the final version is still to come5. Until then, treat the exemption as something to justify in writing rather than assume.

What providers must show about their data

If your organisation builds a high-risk system, or substantially changes one, it is the provider. Article 10 requires training, validation and test data to be governed by documented practices that cover design choices, where the data came from and how it was collected, preparation steps such as labelling and cleaning, the assumptions the data is meant to represent, gaps in the data, and an examination of possible biases with measures to detect, prevent and reduce them3. The data sets themselves must be relevant, sufficiently representative and, to the best extent possible, free of errors and complete for the intended purpose3.

Two further duties land on data platforms. Article 12 requires high-risk systems to record events automatically over their lifetime, so that their operation can be traced, and Article 11 requires technical documentation that describes, among other things, the data the system was built with3. A team that has to reconstruct the lineage of a model trained two years ago will find this much harder than one that records it as it goes.

Using a system brings data duties too

Most organisations will deploy high-risk systems bought from vendors rather than build them. Deployers must use a system according to its instructions, give oversight to people with the competence and authority to intervene, make sure that input data under their control is relevant and sufficiently representative for the intended purpose, and keep the logs the system generates for at least six months3. Employers must inform workers’ representatives before putting a high-risk system to use at work3.

Public bodies, private organisations providing public services, and firms using AI for credit scoring or for life and health insurance pricing must also carry out a fundamental rights impact assessment before first use3. People affected by a decision based on a high-risk system in Annex III can ask for an explanation of the role the system played3. Answering that request depends on the same records: what data went in, which version of the model was running, and who reviewed the result. Breaches of these obligations can bring fines of up to €15 million or 3% of worldwide annual turnover, whichever is higher3.

Standards and guidance are arriving

CEN and CENELEC published EN 18286, a quality management standard for AI Act purposes, in July 2026. It is the first European standard written to support the Act4. Further standards on topics including risk management and logging are still in development. Once a harmonised standard’s reference is published in the EU’s Official Journal, following it gives a presumption that the matching requirements are met, which makes these documents worth tracking even before they are final.

What to do in the next twelve months

  • InventoryList every model and AI feature that touches an Annex III area, including vendor systems, and record whether you are the provider or the deployer.
  • Scope decisionsFor anything you consider outside the high-risk rules, write down why, and review it when the final guidelines are published.
  • LineageFor each in-scope system, record the origin, preparation steps and known gaps of the training, validation and test data.
  • Bias testingAgree how bias is tested, which data is needed to do it, and who signs off the result.
  • LoggingCheck that in-scope systems log events automatically and that the logs can be kept for at least six months.
  • Input dataFor systems you deploy, check that the data you feed them matches the population and purpose the provider designed them for.
  • OversightName the people who oversee each system and make sure they have the training and authority to step in.

We will put these steps to teams that have started at BIAAS 2027 in Amsterdam, in the panel on the EU AI Act in practice.

Sources

  1. Yes, August 2 Still Matters: The EU Approved a High-Risk AI Delay, but Most Transparency Obligations RemainJones Walker, 16 July 2026
  2. Digital Omnibus on AI: Regulation (EU) 2026/1744 Is Published in the Official JournalNicola Fabiano, July 2026
  3. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Articles 6, 10–12, 25–27, 86 and 99, and Annex IIIOfficial Journal of the European Union, 12 July 2024
  4. EN 18286 in the Spotlight: Supporting Compliance with the AI ActCEN-CENELEC, 30 July 2026
  5. Draft EU Guidelines Clarify When AI Systems Are High-Risk Under the AI ActJones Day, June 2026

This report explains the law for planning purposes and is not legal advice.

BIAAS 2027, Amsterdam

Hear it first-hand from the people doing the work

Two days of keynotes, panels and 1-on-1 meetings with senior data, analytics and AI leaders, on 23-24 March 2027.