humanoid4rent
ВойтиПолучить предложения

AI Act harmonised standards: where CEN-CENELEC stands and why the work slipped

EN 18286 became the first harmonised standard under the AI Act in October 2026. The rest of the package is still in drafting or enquiry. Here is what that means for robots.

Автор: Редакция humanoid4rent · опубликовано 10 ноября 2026 г.

Эта страница ещё не переведена
Вы читаете английский текст, потому что версии на языке Русский пока не существует. Ничего не опущено: это вся страница целиком, на том языке, на котором она была написана.

As of November 2026, only one harmonised standard under the EU AI Act has its reference published in the Official Journal: EN 18286, which covers quality management systems. The rest, from risk management to cybersecurity and conformity assessment, are still in drafting or public enquiry. That is why the AI Omnibus moved the high-risk deadlines, and why robot makers cannot yet point to a full set of AI Act standards.

This article explains what harmonised standards do, where the European work stands, why it fell behind, and which questions an event organiser can sensibly put to a supplier today. It is general information, not legal advice.

What a harmonised standard is, and why it matters

EU product laws usually set out requirements in general terms: a system must be robust, must log its activity, must allow human oversight. They do not say how to show that a product meets them. Harmonised standards fill that gap. The Commission asks the European standards bodies to write them, and once a standard's reference appears in the Official Journal, following it gives a legal advantage.

In the AI Act, that advantage is a presumption of conformity. Commission Implementing Decision (EU) 2026/2274 restates the rule: high-risk AI systems that follow harmonised standards whose references are published in the Official Journal are presumed to meet the requirements those standards cover. The words "those standards cover" matter. The presumption extends only as far as the standard does.

Using standards is voluntary. A provider can show compliance in other ways, but it then has to explain and document its own technical solution, which usually takes longer, costs more and is easier to challenge.

Who writes the AI Act standards

The Commission gave the job to CEN and CENELEC, the European standards organisations, which do the work through a Joint Technical Committee called JTC 21. According to the Commission's AI Act standardisation page, the request covers ten areas:

  • risk management
  • governance and quality of datasets
  • record keeping
  • transparency
  • human oversight
  • accuracy
  • robustness
  • cybersecurity
  • quality management
  • conformity assessment

The Commission presents the standards as a way to turn legal requirements into common technical language, giving companies more legal certainty and lowering compliance costs.

Where the work stands in November 2026

The first deliverable is complete. JTC 21 reports that EN 18286 was published in July 2026 and became a harmonised standard in October 2026. The implementing decision citing it is dated 12 October 2026. EN 18286 sets out the quality management system that Article 17 of the AI Act requires from providers of high-risk AI systems, with informative mappings to ISO 9001 and ISO/IEC 42001.

The rest of the package is still moving through the process. In July 2026, JTC 21 listed four drafts in public enquiry, each with its own comment deadline:

  • prEN 18228 on AI risk management, enquiry closing 30 July 2026
  • prEN 18282 on cybersecurity specifications for AI systems, closing 30 July 2026
  • prEN 18229-1, the first part of an AI trustworthiness framework, on logging, closing 20 August 2026
  • prEN 18288, a taxonomy of AI tasks in computer vision, closing 16 July 2026

The same JTC 21 post on EN 18286 says that work on data governance, transparency, human oversight, accuracy and robustness and conformity assessment is still ongoing, and that several core drafts are being revised after public consultation. It also states that the amended standardisation request now runs until 28 February 2027, and calls EN 18286 "the first deliverable" under that request, not the last.

No official source we checked gives firm dates for when each remaining standard will be cited in the Official Journal. A draft passing enquiry is not the same as a harmonised standard: the Commission must still assess it and publish its reference.

Why the work fell behind

The official sources acknowledge the delay without giving a detailed account of its causes. Several factors are on the record.

  1. The scope is wide. Ten requirement areas, several of them without an established international standard to adapt, is a large programme for one committee.
  2. Consensus takes time. JTC 21 works by consensus, which its FAQ defines as the absence of sustained opposition on substantial issues from any important group of interested parties. That process involves national delegations, industry, civil society and public bodies, and it is slow by design.
  3. Drafts stalled. In October 2025, CEN and CENELEC adopted an exceptional package of measures because six drafts had fallen behind. A small drafting group of active experts was set up to finish them, and drafts that pass the enquiry vote can now be published directly, without the usual separate formal vote. CEN and CENELEC said the standards needed to be available by the fourth quarter of 2026, and that enquiry comments would be dealt with in later versions.

The Commission itself named the gap. The explanatory memorandum of its Digital Omnibus on AI proposal of 19 November 2025 points to a lack of harmonised standards for the high-risk requirements, together with delays in common specifications, guidance and national authorities, as a threat to the effective application of those obligations.

The fast-track measures carry a trade-off. Publishing on the strength of an enquiry vote saves months. It also means some stakeholder comments will only be reflected in a later revision, so the first editions of some standards may change.

How the delay changed the AI Act timeline

The proposal would have tied the application of the high-risk rules to the availability of support measures, with fixed backstop dates. According to the Commission's AI Act page, political agreement on the AI Omnibus was reached on 7 May 2026 and it entered into force on 27 July 2026. The high-risk rules now apply from 2 December 2027 for systems in the listed high-risk areas, such as biometrics, employment and critical infrastructure, and from 2 August 2028 for AI systems built into regulated products. The Commission says the aim is for the rules to apply once companies have support tools such as standards.

For robots, the 2028 date is usually the relevant one, because an AI safety function inside a machine falls under the product-linked route. Our AI Act timeline for robots sets out which rules apply when.

If standards are still missing, the Act has a fallback. Under Article 41, the Commission may adopt common specifications by implementing act when harmonised standards are unavailable or insufficient. These specifications carry the same presumption of conformity and are repealed once matching standards are cited. So far the Commission has not announced any common specifications for the high-risk requirements.

What this means for robot makers claiming conformity

For now, a manufacturer or supplier can claim the AI Act presumption of conformity only for quality management, through EN 18286. For the other requirements, they have to rely on their own documented methods, on drafts that may still change, or on international standards that are not harmonised under the AI Act.

Three points help when reading supplier claims:

  • "Compliant with the AI Act" is not yet testable for most robots. The high-risk obligations do not apply until 2027 or 2028, and most of the reference standards are not finished. A claim of full compliance today goes beyond what the framework can support.
  • AI Act standards do not replace machinery safety. Physical risks such as falls, pinch points and collisions are dealt with under machinery and robot-safety rules, which have their own standards. See our pieces on the 2027 Machinery Regulation and on what CE marking does and does not tell you.
  • Using a draft is a choice, not a presumption. A maker that follows prEN 18228 or prEN 18282 is showing good practice, but it gets no legal presumption from a draft.

What this means for event organisers

Most robots hired for events are not high-risk AI systems in the AI Act sense, and the delay does not change the safety checks you need on the day. It does tell you what kind of evidence to expect.

  1. Ask suppliers for the robot's CE declaration of conformity and which directives and standards it lists, rather than accepting a general claim of AI Act compliance. Add this to your brief to suppliers.
  2. If a supplier mentions EN 18286, understand that it concerns the manufacturer's quality management, not the robot's behaviour around a crowd.
  3. Keep on-site safety planning, such as exclusion zones, operators and stop procedures, separate from the paperwork; our public event safety guide covers this.
  4. Check liability and insurance cover in the contract, because standards status does not settle who pays if something goes wrong. See insurance and liability for robot hire.
  5. Expect this picture to change during 2027 as more drafts reach the Official Journal.

Frequently asked questions

Which AI Act harmonised standards have been published so far?

As of November 2026, one: EN 18286:2026 on quality management systems for AI Act purposes. Its reference was published through Commission Implementing Decision (EU) 2026/2274, dated 12 October 2026. Drafts on risk management, cybersecurity, logging and other topics have been through public enquiry or are still in drafting, and none of them yet gives a presumption of conformity.

What is a presumption of conformity?

When a product follows a harmonised standard whose reference is published in the Official Journal, authorities presume it meets the legal requirements that standard covers. The presumption only goes as far as the standard. A manufacturer can still use other methods, but must then document and justify them itself, which is usually harder to defend.

Why were the AI Act standards delayed?

Official sources confirm the delay but give few details. The request covers ten requirement areas, CEN-CENELEC JTC 21 works by consensus across many stakeholders, and CEN and CENELEC said in October 2025 that six drafts had fallen behind. They introduced fast-track measures, including direct publication after a positive enquiry vote, to catch up.

What happens if the standards are not ready in time?

Article 41 of the AI Act lets the Commission adopt common specifications by implementing act when harmonised standards are missing or insufficient. These give the same presumption of conformity and are withdrawn once matching standards are cited. The AI Omnibus also moved the high-risk rules to 2 December 2027 and 2 August 2028.

Does EN 18286 make a robot safe to use at an event?

No. EN 18286 deals with how a provider of high-risk AI systems organises its quality management system. It says nothing directly about how a robot behaves near visitors. Physical safety is covered by machinery and robot-safety rules, your risk assessment, trained operators and the site measures agreed with the venue.

How this article was made: written by the humanoid4rent editorial team with AI assistance; every factual statement was checked against the sources listed below on 2026-11-10. Spotted an error? Write to info@humanoid4rent.com.

Источники

Факты в этой статье были проверены по данным страницам в день её написания.

  1. Standardisation of the AI Act – European Commission, Shaping Europe's digital future
  2. Update on CEN and CENELEC's Decision to Accelerate the Development of Standards for Artificial Intelligence – CEN-CENELEC
  3. Commission Implementing Decision (EU) 2026/2274 on the harmonised standards for high-risk AI systems drafted in support of Regulation (EU) 2024/1689 – EUR-Lex
  4. EN 18286 becomes the first harmonised standard under the EU AI Act – CEN-CENELEC JTC 21
  5. Significant milestone for European AI standardization – CEN-CENELEC JTC 21
  6. Proposal for a Regulation amending Regulation (EU) 2024/1689 (Digital Omnibus on AI), COM(2025) 836 final – EUR-Lex
  7. AI Act – European Commission, Shaping Europe's digital future
  8. Article 41: Common specifications – AI Act Service Desk, European Commission

Ещё в блоге

Хотите видеть робота на своём следующем мероприятии?

Один раз опишите дату, место и задачи робота. Поставщики ответят предложениями, которые вы сможете сравнить по одинаковым условиям.

Отправить заявку

Работаете в робототехнике? Пишите для нашего блога. →