When does the EU AI Act apply to our systems?
The EU AI Act applies in stages. Chapters I and II, covering scope, definitions, AI literacy and prohibited practices, have applied since 2 February 2025, and most of the rest of the Act applied from 2 August 2026. The Digital Omnibus on AI, Regulation (EU) 2026/1744, moved the high-risk duties for Annex III systems to 2 December 2027 and for Annex I product systems to 2 August 2028. If your roadmap still shows August 2026 for high-risk obligations, it predates the amendment and needs to be re-dated against the consolidated text of the Regulation.
The date that applies to a given system depends on three things: your role (provider, deployer, importer or distributor), the risk tier the system falls into, and whether the system was already on the market before the relevant date. A prohibited practice carries no transition at all. Transparency duties under Article 50 follow their own dates, and some of them carry grace periods for generative systems placed on the market before 2 August 2026. Legacy high-risk systems come into scope only after significant design changes, with longer deadlines for systems used by public authorities. Confirm each of these points against the consolidated text on EUR-Lex rather than relying on secondary summaries.
Start with an inventory: list every AI system you build, buy or deploy, record your role for each, and tag its likely risk tier. Then map each entry to the dates in this timeline and note whether it was placed on the market before or after each cutoff. Ask your legal counsel to confirm the classification of borderline systems, because the answer determines whether a 2027 or 2028 deadline applies, or whether the system falls outside high-risk scope altogether. Record the reasoning, since a market surveillance authority will ask how you reached each date.
What already applies and what it means today
Three application dates have already arrived, and together they cover the scope rules, the AI literacy duty, the prohibited practices, the general-purpose AI rules, governance, penalties and the general application of the Act. Each card below explains what the date switched on and what an organization should check today. National authority designations vary by member state, so confirm them locally.
Scope and definitions apply
Since 2 February 2025, the scope and definitions in Chapters I and II have applied, so the Act's vocabulary governs how your systems are classified. Start by confirming whether you are a provider, deployer, importer or distributor for each AI system you build or use. Record that role in your AI system inventory, because the duties that follow depend on it.
AI literacy duty under Article 4
Providers and deployers must take measures to ensure a sufficient level of AI literacy among their staff and others who operate AI systems on their behalf. The duty applies from 2 February 2025 and has no fixed training format. Check who operates each system, what they already know, and whether your training records show that the measures were taken.
Prohibited practices under Article 5
The prohibited practices in Article 5 have applied since 2 February 2025, and they are the only AI uses that the Act bans outright. Review existing systems against each prohibition, including any emotion recognition or social scoring features bundled into vendor products. Ask vendors to confirm in writing which prohibitions their products cannot trigger.
General-purpose AI and governance rules
Since 2 August 2025, the rules on general-purpose AI models, governance, notified bodies and penalties have applied, except Article 101. Member States had to designate national competent authorities by that date, so confirm which authority supervises your activity in each country where you operate. If you build on a general-purpose model, document which provider's model you use and what that provider has made available to you.
General application from 2 August 2026
From 2 August 2026 the Act applies generally, including the Article 50 transparency duties for chatbots, synthetic content and emotion recognition. The Commission also gained the power to fine general-purpose AI providers under Article 101 from this date. Check every customer-facing AI interaction for a disclosure, and confirm with your national authority how transparency is supervised.
How the application dates run from December 2026 to 2030
The sequence shows which AI Act duties start in each phase, what each phase produces for a compliance plan, and which dates depend on guidance and standards that are still arriving. Organizations use it to decide what must be ready now and what can wait for the 2027 and 2028 high-risk dates.
Prepare for the December 2026 prohibition and marking duty
From 2 December 2026, the new prohibition on non-consensual intimate imagery and child sexual abuse material applies, and providers of generative systems placed on the market before 2 August 2026 must meet the Article 50(2) marking duty. Product and engineering teams should confirm which generative systems they provide, whether those systems were placed on the market before 2 August 2026, and how output is marked. The output is a dated register of affected systems and a marking method each provider can show to an assessor.
Confirm general-purpose AI model compliance by August 2027
Providers of general-purpose AI models placed on the market before 2 August 2025 must comply by 2 August 2027. Legal and compliance leads should identify any such model they supply, compare its technical documentation with the Chapter V duties, and decide whether the Code of Practice will be used as a route to show compliance. The output is a documentation file per model and a named owner for the gap list.
Plan participation in a regulatory sandbox
By 2 August 2027, each Member State must have at least one AI regulatory sandbox operational. Organizations developing innovative systems should check which national authority runs the sandbox in each Member State where they operate and whether the application process is open. The output is a short list of target sandboxes and a decision on whether to apply, based on the national authority's published terms.
Track Commission guidance due in 2027
The Commission must issue guidelines on integrating AI Act and product-law compliance by 1 August 2027, and post-market monitoring guidance with a template by 2 September 2027. Compliance leads should assign an owner to monitor these publications and compare them with the post-market monitoring plan and product-law obligations already in place. The output is a dated review of each guidance document and a list of plan changes it requires.
Track notified body designation applications due January 2028
Notified bodies designated under Annex I Section A product legislation must apply for AI Act designation by 28 January 2028. Manufacturers of products that rely on these bodies should ask their current notified bodies whether they have applied and how the conformity assessment timeline will be affected. The output is a written confirmation from each relevant body, or a documented decision to seek another body.
Meet the Annex III high-risk rules from December 2027
The high-risk rules in Chapter III Sections 1 to 3 apply to Annex III systems from 2 December 2027. Providers and deployers should finish the Articles 8 to 15 requirements, quality management, technical documentation and registration for these systems before that date, using the conformity route that applies to each system. The output is a signed conformity declaration or internal control file for each Annex III system, plus deployer records for systems in use.
Meet the Annex I product high-risk rules from August 2028
From 2 August 2028, the high-risk rules apply to AI systems that are safety components of Annex I products. Manufacturers should combine AI Act evidence with the conformity procedure of their product legislation and confirm whether a notified body is required. The output is a single conformity file per product that covers both the AI Act and product-law requirements.
Close legacy and public authority obligations by 2030
High-risk systems already on the market before the application dates come into scope only after significant design changes, but systems intended for public authorities must comply by 2 August 2030. Annex X large-scale IT system components must comply by 31 December 2030. Public bodies and operators of legacy systems should keep a log of design changes and check whether each system is intended for a public authority. The output is a list of legacy systems with the trigger that will bring each one into scope.
Why the high-risk dates moved and what that does not change
The Digital Omnibus on AI moved the high-risk dates, but it did not change what a high-risk system must do. Regulation (EU) 2026/1744 amended the AI Act on 8 July 2026, was published on 24 July 2026 and entered into force on 27 July 2026. The Commission had proposed it on 19 November 2025 as COM(2025) 836, and the institutions reached political agreement on 7 May 2026. Under the amendment, the Chapter III Sections 1 to 3 rules for Annex III systems now apply from 2 December 2027 instead of 2 August 2026, and for Annex I product systems from 2 August 2028 instead of 2 August 2027. The regulation itself is the primary source for these dates: consolidated text of Regulation (EU) 2024/1689 as amended in 2026.
The recitals attribute the delay to two causes: harmonised standards, common specifications and guidance arrived late, and national competent authorities were established late. Those reasons matter for planning because the requirements still depend on the standards that describe how to meet them. The first harmonised standard under the Act, EN 18286:2026 on quality management, was cited in the Official Journal on 14 October 2026 under Commission Implementing Decision (EU) 2026/2274. A quality management system built to that standard is one way to support the presumption of conformity, so teams should check whether their chosen approach maps to it and confirm with their national authority how it will be assessed.
What the delay does not change is the content of the obligations. Articles 8 to 15 still define what a high-risk system must demonstrate, and providers and deployers still owe their documentation, logging, human oversight and monitoring duties once the rules apply. The delay only gives organizations more time to prepare. Treat the new dates as a planning window, not a reason to pause work, and ask your legal team and any implementation partner to confirm in writing which transition rules apply to a system already on the market, since the amended Article 111 sets separate deadlines for those systems.
Transition periods for legacy and existing systems
High-risk AI systems placed on the market or put into service before the high-risk application date are not automatically pulled into the new rules on that date. Under Article 111, they come into scope only when they undergo significant changes in their design. An update that is routine maintenance or a security patch does not, on its face, change the design, but a change to the model, intended purpose or core architecture usually does. Whether a particular change is significant is a legal and technical judgment that should be documented and reviewed with counsel.
- Placement date: Record the date each system was first placed on the market or put into service, not the date you first deployed it internally.
- Design change log: Keep a dated record of model retraining, changes to intended purpose and architectural changes, with the reasoning for each significance decision.
- Scope trigger: Once a system is substantially modified, treat it as in scope for the obligations that apply on the date of that change.
- Ongoing duties: Legacy status does not remove existing obligations under other laws, such as data protection or sector rules.
- Documentation of the decision: Keep the reasoning behind each scope call so an assessor can follow it.
Check each system's placement date and change history against the Article 111 wording with counsel before you rely on legacy treatment.
Article 111 gives a longer runway for high-risk AI systems intended for use by public authorities. These systems must comply by 2 August 2030, which is later than the general high-risk dates. The extension is aimed at public bodies that may need more time to procure, migrate and document systems in use. It does not exempt a public authority from the Act for systems placed on the market after the general dates, and the boundary between a public authority system and a commercial one depends on intended use.
- Intended use test: Decide whether the system is intended for public authorities. A commercial product sold to a government body can still be a commercial system placed on the market.
- Deadline: Plan for 2 August 2030 as the outer date for compliance of qualifying systems.
- Procurement terms: Ask suppliers for their position on which date applies to the product and whether an update would move it into scope earlier.
- Interaction with other laws: Public sector duties under national administrative and data protection law continue to apply in parallel.
- Migration planning: Factor the replacement or upgrade cycle of public systems into the timeline, since the extension has an end date.
Confirm with counsel whether each public deployment qualifies and how the intended use is recorded in procurement files.
Article 111 also sets a separate date for components of large-scale IT systems listed in Annex X. These components must comply by 31 December 2030. Annex X covers large-scale IT systems used in the area of freedom, security and justice, such as shared EU databases operated by public bodies. Most organizations will not operate these systems directly, but a supplier, integrator or contractor working on one should confirm whether the component it provides is covered.
- Identify the component: Map which parts of your architecture form a component of an Annex X system rather than a standalone AI system.
- Confirm the Annex X listing: Check the annex text directly, since the scope depends on the listed systems rather than on the technology used.
- Date of compliance: Use 31 December 2030 as the outer date for qualifying components.
- Supplier questions: Ask integrators which compliance date they assume for the components they deliver.
- Change tracking: Record any modification to a covered component, since changes can affect the analysis.
Confirm the Annex X scope and the component boundary with counsel, because the date depends on the classification of each part of the system.
Providers of general-purpose AI models placed on the market before 2 August 2025 have until 2 August 2027 to comply with the general-purpose AI obligations under Article 111(3). This transition applies to the model, not to every product built on it. Organizations that fine-tune, host or build on such a model should check whether they have become a provider of their own model or system, because that status determines which duties apply and when.
- Placement date of the model: Confirm when the model was first placed on the EU market, since the 2 August 2025 cut-off depends on it.
- Transition deadline: Plan for 2 August 2027 as the compliance date for qualifying models.
- Downstream systems: A system built on a legacy model still has its own placement date and its own high-risk analysis.
- Supplier documentation: Request the provider's compliance documentation and its position on the transition date.
- Role check: Substantial modification or rebranding can change who is the provider.
Confirm the placement date of each model with its provider and with counsel before relying on the 2027 date.
Providers of generative AI systems placed on the market before 2 August 2026 must meet the marking duty in Article 50(2) from 2 December 2026, according to the transition provision in Article 111(4). Article 50(2) requires that synthetic audio, image, video or text output be marked in a machine-readable way and be detectable as artificially generated or manipulated. The transition gives legacy generative systems a short window to add these markings, while systems placed after the cut-off must meet the duty from the outset.
- Placement date: Record when each generative system was first placed on the market, since the transition covers only systems placed before 2 August 2026.
- Marking deadline: Plan for 2 December 2026 for the Article 50(2) marking duty on legacy systems.
- Technical scope: Confirm which output types your system generates, since the duty covers synthetic content of several kinds.
- Code of practice: Review the Code of Practice on Transparency of AI-generated Content, published on 10 June 2026, as a practical reference.
- Supplier evidence: Ask model and platform providers how their outputs are marked and how detection works.
Confirm with counsel whether a given system counts as placed on the market before the cut-off, and how the marking duty applies to its outputs.
What to have in place before each date
The AI Act timeline turns into work in a few clear steps: know which AI systems you run, decide which roles you hold, classify each system, and build the evidence before the high-risk dates arrive. The organization owns every one of these controls. A partner can prepare systems, documentation and records, but only an independent auditor or notified body issues a conformity outcome.
System inventory and role mapping
Before 2 December 2026, list every AI system and general-purpose AI model you build, buy, fine-tune or deploy, including shadow tools used by business teams. For each entry, record whether your organization acts as provider, deployer, importer or distributor, and which vendor supplies the model. Confirm who owns each entry and where its logs and contracts sit. Without this list, no later step about classification, documentation or retention can be scoped.
Classification and gap assessment
Before 2 December 2027, classify each inventoried system against the prohibited practices, the Annex III categories and the Annex I product route, and document the reasoning, including any Article 6(3) exemption claim. Then compare current practice with the Chapter III requirements for each high-risk system. Ask your counsel or assessor to confirm the classification, since the Commission's draft high-risk guidelines are not final.
Quality management and technical documentation
For Annex III systems, put a quality management system and technical documentation in place before the system is placed on the market or put into service. Assign owners for risk management, data governance, logging, human oversight and post-market monitoring. Confirm which parts your organization will build internally and which a partner will help prepare. The first harmonised standard on quality management, EN 18286:2026, was cited in the Official Journal on 14 October 2026, so check it against your process.
Conformity route and notified body need
For Annex I product systems, which apply from 2 August 2028, establish early whether the product's own legislation requires a notified body. Notified bodies designated under Annex I Section A product legislation must apply for AI Act designation by 28 January 2028, so check the capacity and scope of your candidate body well before then. Most Annex III systems follow internal control, but some biometric systems do not. Confirm the route with the independent assessor you plan to use.
Evidence retention and records
Decide now how long you keep logs, technical documentation, training data records and incident reports, and who can retrieve them. Documentation must stay available to national market surveillance authorities for the period the Act requires, so confirm that period against the current consolidated text. Set access controls and tamper evidence on the systems that hold these records, and test that you can produce a complete evidence set for one system on short notice.
Who owns the timeline work
Your organization owns the timeline. Only your management can decide which AI systems are in scope, which role you hold for each one and how each system maps to the application dates in the amended Act. Every later step depends on those decisions being recorded and assigned to named people.
- System inventory: Keep a register of every AI system you build, buy or deploy, with its intended purpose, supplier and EU use. A date cannot be applied to a system you have not listed.
- Legal and role decisions: Decide whether you act as provider, deployer, importer or distributor for each system, and whether it falls into a high-risk route. Counsel should record the reasoning for each decision.
- Deadline ownership: Assign one owner to each date, including 2 December 2026, 2 December 2027, 2 August 2028 and 2 August 2030 where legacy public authority systems are concerned.
- Supplier contracts: Ask each supplier which role it plans to hold, which dates it is planning for and which documentation it will hand over. Read the official text of the consolidated Regulation before you draft these clauses.
- Governance sign-off: Have executive leadership approve the plan, the owners and the evidence your organization must keep.
Check first that your inventory lists every system, because every other date in the plan depends on it.
An implementation partner such as DeployOpen works inside your environment to prepare systems and evidence against the plan you have agreed. DeployOpen plans and implements self-hosted open-source platforms in your own cloud accounts, data centers or air-gapped networks, then prepares the technical and documentation evidence an assessor will later review. It does not decide your legal role for any system, and it does not issue reports, attestations or certificates.
- Logging and records: Configure logging, retention and access records for high-risk systems, scoped to the retention rules your legal team sets.
- Technical documentation: Draft technical documentation and risk management records from your system inventory, for your review and approval.
- Quality management evidence: Prepare procedures, change records and test results that an independent assessor can sample.
- Fixed-scope delivery: Work runs to a written scope with agreed deliverables and timelines, settled in writing before work starts.
- Operation after go-live: Retainers can cover upgrades, monitoring and continuous testing when your plan runs past the first application date.
Check that the written scope names the systems, dates and evidence included, and that it states which decisions remain with your organization.
Where the Act requires an outside body, an independent assessor or notified body makes the conformity decision. Most Annex III high-risk systems use internal control under Annex VI, so the provider's own assessment is the route for them. Notified bodies are involved for some biometric systems and for Annex I products under Article 43, and notified bodies designated under Annex I Section A product legislation must apply for AI Act designation by 28 January 2028.
- Conformity outcomes: Notified bodies issue certificates where the route requires them. No partner can stand in for that decision.
- Route by system: Confirm with the assessor whether each system follows internal control or a notified body route under Article 43.
- Lead times: Ask about availability early, since designation deadlines and demand for assessment capacity may affect your schedule.
- Independence: Keep the preparation team and the assessment team separate in the contract so the assessor's judgment stays independent.
- Standards: Ask how the assessor reads harmonised standards. EN 18286:2026 on quality management was cited in the Official Journal on 14 October 2026.
Check the assessor's designation or accreditation status for your system type before you sign any engagement.
Questions buyers ask about AI Act dates
Which AI Act date applies to our system?
The applicable date depends on the system's risk tier and your role, so classify the system first and then match the tier to its dates. Article 5 prohibitions and the Article 4 AI literacy duty have applied since 2 February 2025. The general-purpose AI model rules have applied since 2 August 2025. Most of the Act, including the Article 50 transparency duties, applies from 2 August 2026. Generative systems placed on the market before that date must meet the Article 50(2) marking duty from 2 December 2026. High-risk systems in Annex III apply from 2 December 2027, and Annex I product systems from 2 August 2028. Significant design changes can bring a legacy system into scope, so confirm your classification, role and dates with counsel and your national market surveillance authority.
Is the 2026 delay for high-risk AI systems final?
Yes, the delay is in law. Regulation (EU) 2026/1744 amended the AI Act on 8 July 2026, was published on 24 July 2026 and entered into force on 27 July 2026. It moved the Annex III high-risk date to 2 December 2027 and the Annex I date to 2 August 2028. The recitals attribute the delay to late availability of standards, common specifications and guidance, and to the late establishment of national competent authorities. Guidance is still arriving. The Commission owes guidelines on integrating AI Act and product-law compliance by 1 August 2027, and post-market monitoring guidance with a template by 2 September 2027. The first harmonised standard, EN 18286:2026 on quality management, was cited in the Official Journal on 14 October 2026. Read the consolidated text as amended in 2026 and confirm with counsel whether any further change affects your plan.
What happens if we miss the 2 December 2026 date?
Nothing in the high-risk timeline falls due on 2 December 2026. That date brings the new prohibition on non-consensual intimate imagery and child sexual abuse material, and the Article 50(2) marking duty for providers of generative systems placed on the market before 2 August 2026. Breaching a prohibition can lead to fines of up to EUR 35 million or 7% of worldwide annual turnover, whichever is higher, under Article 99. National market surveillance authorities enforce the Act for AI systems, so the practical exposure includes regulator inquiries, corrective orders and contract disputes with customers who depend on your compliance. How any authority would act depends on its own procedures and the facts of each case, so ask counsel to map your exposure to the specific systems you run.
How long does AI Act preparation take for our organization?
No single duration applies, because the time depends on what you must prepare and how much evidence you already hold. The main drivers are the number of AI systems in the inventory, whether any system is high-risk under Annex III or Annex I, and whether a system needs a notified body. Other drivers include how many suppliers and deployers sit in your value chain, how well your existing ISO 27001, ISO/IEC 42001 or quality management records map to the Articles, and how quickly your teams can produce logs, documentation and human oversight records. Harmonised standards and Commission guidance are still being published, which can change the work. Ask your counsel and consultants to show their assumptions against the 2027 and 2028 dates, and treat any duration they quote as an estimate to test against your inventory.
Can we prepare for the AI Act in-house?
Yes, most organizations keep ownership in-house, and the Act places duties on the provider or deployer rather than on an outside firm. Your organization owns its controls, its risk classification and its decisions on whether a system is compliant. Independent assessors and notified bodies are a different matter: most Annex III systems use internal-control conformity assessment under Annex VI, while notified bodies are involved for some biometric systems and Annex I products under Article 43. An implementation partner such as DeployOpen can prepare systems, controls and evidence, but it does not issue reports or certificates. In-house teams usually need outside help on technical logging, testing, security evidence and documentation of self-hosted platforms, so decide which of those you can staff before you plan the program.
What should we ask a partner for in a fixed-scope quote?
A fixed-scope quote should define the work precisely enough that both sides can see what is included and excluded before anything starts. Ask the partner to state:
- The AI systems in scope, with the classification and role your organization holds for each.
- The dates the plan targets, including 2 December 2026, 2 December 2027 and 2 August 2028 where they apply.
- The deliverables, such as the inventory, technical documentation, logging design, human oversight procedures and evidence packs, and who signs each one off.
- Exclusions, especially any conformity assessment, audit, notified body fee or certification, which an independent body must perform.
- Assumptions about the environment, the systems you already run, and the information you must supply.
- The change process for new scope and any retainer for later upgrades or testing.
Fixed scope gives a fixed basis for comparing offers, so check that each quote uses the same classification and dates.
Official sources to check each date against
Before you commit a compliance calendar to any date in this timeline, check it against the primary text. The most reliable starting point is the consolidated text of Regulation (EU) 2024/1689 as amended by Regulation (EU) 2026/1744, which shows the Act as it now reads after the 2026 changes. Read the application dates in Article 113 and the transitional provisions in Article 111 in that version, not in the 2024 original, because several dates moved after the Act was first published. Ask your legal or compliance team to record the consolidated version date next to every deadline in your plan so that a later amendment can be traced.
The amending instrument itself is Regulation (EU) 2026/1744, the Digital Omnibus on AI. Read its articles, not only summaries, to confirm which provisions apply from 27 July 2026, which take effect on later dates, and which transitional rules apply to systems already on the market. The recitals explain the reasons for the high-risk delay, which helps when you explain the change to a board or an auditor. If your organization relies on a supplier's statement about what changed, match that statement against this instrument and record any gap.
For guidance, the Commission's AI Act policy page on Shaping Europe's digital future is the place to find current guidelines, codes of practice, standards and implementation news. Guidance is not law, and it can be updated or replaced, so note the publication date of any guideline you rely on and recheck it before each review cycle. Assign one named owner in your organization to monitor this page and the Official Journal, and to log any change that affects your application dates.
How DeployOpen plans AI Act readiness around these dates
DeployOpen starts from the application date that applies to each of your AI systems, then plans the platforms, controls and evidence needed before that date. We prepare systems and documentation inside your own infrastructure, while your organization owns its obligations and an independent assessor or notified body issues any formal outcome.
How we scope
- We map each AI system in your inventory to the application date that applies to its role and risk tier.
- Written scope sets which systems, documents and evidence are covered before any work starts.
- Engagements are fixed-scope, with deliverables and timelines agreed in writing, and retainers are available for ongoing support.
How we build
- We deploy and configure self-hosted open-source platforms inside your cloud accounts, data centers or air-gapped networks.
- We set up logging, record keeping and access controls so that evidence is generated as systems run.
- We run security testing of applications, APIs and AI systems to check the controls you rely on before each deadline.
How we hand over
- You receive documentation, configuration records and a dated plan that tracks each application date against your systems.
- We document how each control and piece of evidence maps to the requirement it supports, so your assessor can test it.
- Ownership of controls, policies and sign-off stays with your organization, and we confirm handover in writing.
How we support
- Retainers cover upgrades, monitoring of Commission guidance changes and continued testing after the first deadline.
- We can review the plan when the Commission publishes new guidelines, standards or national authority decisions that affect your dates.
- Formal conformity assessment, certificates and reports come from independent auditors or notified bodies, which we can help you prepare for.

