EU AI Act Deadline Checker

THE DISTINCTION MOST COMMENTARY GETS WRONG: the high risk chapter was deferred. The Act was not.

Quadrant Three // When Agents Rule

EU AI Act Deadline Checker

Answer questions about one system and get the dates that apply to it after the 2026 amendments, with what was deferred separated from what was not, the maximum exposure, and a board paper.

What moved Chapter III Sections 1, 2 and 3: Annex III high risk obligations to 2 December 2027, Annex I embedded high risk to 2 August 2028. National regulatory sandboxes to 2 August 2027.
What did not move Article 5 prohibitions. Article 4 AI literacy, now worded as measures to promote it. Chapter V general purpose AI. Article 50 transparency. Article 49 registration. The general application date of 2 August 2026.
9 questions 5 minutes 1 system per run 0 answers sent anywhere
This is not legal advice and it cannot be. Classification under Article 6 is a documented legal assessment that belongs to qualified counsel, and this tool cannot see your system. It applies the published date structure to your answers, cites the article behind every output so counsel can check it, and records a verification field. Treat the result as a structured question to take to legal. The penalty figures are statutory ceilings, not a forecast of any fine.

Why it exists anyway. The most expensive error available in this area is reading “the AI Act was delayed” as “nothing applies yet”. Several obligation sets are live today, and more arrive before the deferred high risk dates. This tool separates them for one named system.

Your answers are computed in your own browser and are not transmitted or stored.
i

The system and your position

Your role changes which duties attach, and the placement date changes which transitional rules apply.

Three fields here change the answer materially. Article 50 splits duties between providers and deployers. The placement date decides whether the Article 111 transitional rules apply, including whether a high risk system already on the market is caught at all. And SME or small mid cap status changes the penalty ceiling.
EUR

What size is the undertaking for the purposes of the Regulation?

For SMEs and start ups every fine is capped at the lower of the fixed sum and the percentage. Since the 2026 amendments, small mid cap enterprises get the same lower cap for operator obligations and misleading information, but not for prohibited practices. Article 99(6), 99(6a), Article 3(14a), 3(14b)

Is the system intended to be used by public authorities?

High risk systems intended for public authorities must comply by 2 August 2030 even if they were on the market before the high risk rules applied. Article 111(2)

1

The prohibited check, first

This comes before the high risk question, because a prohibited practice cannot be made compliant.

Does the system do any of the things Article 5 prohibits?

Including: subliminal or manipulative techniques that materially distort behaviour; exploiting vulnerability by age, disability or social or economic situation; social scoring; predicting criminal offences based on profiling alone; untargeted scraping to build facial recognition databases; inferring emotions in the workplace or in education, other than for medical or safety reasons; biometric categorisation to infer protected attributes; and real time remote biometric identification in public for law enforcement outside the narrow exceptions. From 2 December 2026 the list also covers systems generating non consensual intimate imagery of identifiable people and child sexual abuse material. Article 5

2

Is a general purpose model involved, and whose is it?

Chapter V attaches at the model layer and has applied since August 2025. It was not deferred.

What is your relationship to any general purpose AI model behind this system?

A provider of the model carries Chapter V obligations directly. Building a system on somebody else’s model does not transfer them to you, but it leaves you with the Article 50 duties at the system layer. Articles 53, 55, 101, 111(3)

3

The Annex I product route

Both conditions must hold, and the 2026 amendments narrowed both.

Is the system a safety component of, or itself, a product covered by the Annex I harmonisation legislation, and does that product require third party conformity assessment for health and safety risks?

Annex I covers machinery, medical devices, in vitro diagnostics, lifts, toys, radio equipment, pressure equipment, civil aviation, vehicles, marine equipment, rail and more. Since 27 July 2026 a system used solely for user assistance, performance optimisation, service efficiency, automation, convenience or quality control not related to safety is not a safety component, and third party assessment required only for risks other than health and safety does not meet the second condition. Article 6(1), 6(1a) to 6(1c)

4

The Annex III use case route

Eight areas. Being listed does not automatically make a system high risk, but it moves the burden to you.

Does the system operate in any of the Annex III areas?

Choose the closest. Where more than one applies, choose the one with the greatest effect on individuals. Article 6(2), Annex III

5

The Article 6(3) filter

Answer this even if you chose none above, because the answer is checked against the area question.

The filter is a regulated act, not a private judgement. Claiming it requires a documented assessment under Article 6(4), which applies with the Annex III obligations from 2 December 2027, and registration in the EU database under Article 49(2), which the amendments kept in simplified form and which sits in the part of the Act applying from 2 August 2026. It never applies where the system performs profiling of natural persons, which is the condition most often missed.

Which best describes what the system does within that area?

The filter requires that the system does not pose a significant risk of harm, including by not materially influencing the outcome of decision making, and that at least one of four conditions is met. Article 6(3), 6(4), 49(2)

6

Transparency triggers

Live since 2 August 2026 and not deferred. The obligation set most often missed by organisations that believed the Act had been delayed.

Which of these does the system do? Tick every one that applies.

Duties split by role. Making people aware they are interacting with AI and machine readable marking of synthetic content fall on the provider. Informing people subject to emotion recognition or biometric categorisation, and disclosing deep fakes, fall on the deployer. Article 50(1) to 50(4)

What has your legal function actually confirmed?

This drives the confidence grade rather than the dates. A classification nobody has reviewed is a working assumption.

0 of 9 answered

The ask

The decision, the classification, the live obligations, the exposure and the owner, in the order a board wants them. Everything below is the evidence.

Indicative classification

Minimal risk

What was deferred, and what was not

The distinction this tool exists for, applied to your system rather than in general.

Live or arriving regardless

Dates not moved by the 2026 amendments

    Deferred by the amendments

    Later than originally enacted

      The dates that apply to this system

      Filtered to your answers. Verify every row against the consolidated text before relying on it.

      Swipe the table sideways to see every column.

      DateWhat appliesArticleDeferredDaysStatus

      Maximum penalty exposure

      Article 99 and 101 ceilings applied to your turnover. Statutory maxima in EUR, not expected outcomes.

      Your duties, by role

      The same system produces different obligations for different parties in the chain. These are yours.

        Do your own answers agree with each other?

        Contradictions are answers that cannot both be true. Findings and caveats are consistent answers that still change what you must do or how far the reading can be relied on.

        Confidence in this reading

        Graded on whether the classification would survive a market surveillance authority asking to see it.

        ?

        Not graded

          The reading a regulator would take

          Article 80 allows a market surveillance authority to evaluate a classification and require compliance where it considers the system misclassified. This is the worse reading your answers leave open.

          Same

          classification if an open answer goes against you

          None

          the date that would then apply

          Board paper

          Written to be pasted into a board, risk or audit committee paper. Complete the bracketed fields, and have counsel confirm the classification before it is circulated.

          Turn the dates into work

          A date is a diagnosis. These are the instruments that meet it.

          Track every obligation, not one system

          The Regulatory Obligation Register carries dated obligations across the AI and cryptographic regimes, each with a confidence rating and a verification field, plus a per system register with escalation flags. Rebuilt in September 2026 for the amended timeline.

          Get the register

          Produce the documentation

          Article 11 and Annex IV technical documentation, and the Article 6(4) assessment behind a filter claim, both need an inventory underneath them. The Agent Inventory workbook records classification per system, and the AIBOM Readiness Checklist scores whether you could produce the file on request.

          Get the instruments

          Understand the fragmentation

          Chapter 7 of When Agents Rule covers the regulatory stack and why compliance in one jurisdiction can create exposure in another. Chapter 8 covers transparency and documentation. Chapter 14 covers operating across more than one regime at once.

          Get the book

          The deferral was not a reprieve

          Prohibitions, AI literacy, general purpose AI, transparency and registration are live today.

          The book page Free toolkit Get the book Advisory services When Clouds Fail

          This is not legal advice, and it is not capable of being legal advice. This tool applies a published date structure to answers supplied by the user. It does not determine legal status, it cannot inspect the system, and classification under Article 6 of Regulation (EU) 2024/1689 is a documented legal assessment that belongs to qualified counsel. Article 80 permits a market surveillance authority to evaluate a classification and to require compliance where it considers a system misclassified.

          Sources and currency. Dates reflect Article 113 and Article 111 of Regulation (EU) 2024/1689 as amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026 and in force from 27 July 2026. The amendments deferred Chapter III Sections 1 to 3 (other than Article 6(5)); they did not defer Chapter IV or Chapter III Section 5, which includes registration under Article 49. Article 49(2) registration for systems claimed under Article 6(3) was retained in simplified form. Penalty ceilings reflect Articles 99 and 101, including new Article 99(6a) for small mid cap enterprises. This build is current as at 16 September 2026. Dates in this area have moved once and may move again, so verify every row against the Official Journal and your own counsel before acting. This is not legal, technical or financial advice. Your answers are computed in your browser and are not transmitted or stored by this tool.