EU AI Act Timeline After the Omnibus

(updated: ) · 14 min read · Lyudmil Arkov

This is the EU AI Act timeline after the Omnibus. The Act (Regulation (EU) 2024/1689) is in force. Prohibited practices have applied since February 2025, and GPAI rules since August 2025. The Digital Omnibus on AI (Regulation (EU) 2026/1744, in force since 27 July 2026) moved the high-risk dates: 2 December 2027 for Annex III systems, and 2 August 2028 for Annex I systems.

Two further prohibitions apply from 2 December 2026. Article 2 covers providers who place a system on the market or put it into service in the Union, providers who place a general-purpose AI model on the Union market, and deployers established or located in the Union. It also covers providers and deployers outside the Union whose output is used in the Union. Importers, distributors, and product manufacturers have duties too. Most of your systems aren’t high-risk. The dates below are the ones to plan against.

What does the EU AI Act regulate?

The regulation takes a risk-based approach. Instead of treating all AI the same, it classifies AI systems into four risk tiers - each with different obligations. Think of it as a proportional framework: the higher the risk your AI poses to people’s health, safety, or fundamental rights, the more you need to do.

This is actually sensible design. Your internal meeting summarizer doesn’t need the same scrutiny as an AI system deciding who gets a loan. The EU AI Act recognizes that, and so should your compliance strategy.

How does risk classification work?

Here’s the breakdown:

  • Unacceptable risk (banned) - Article 5(1)(a) to (h) have applied since 2 February 2025. In short, they cover harmful manipulation, exploitation of vulnerabilities, social scoring by public or private actors, assessing or predicting the risk of a criminal offence based solely on profiling or on personality traits, untargeted scraping of facial images, inferring emotions in the workplace and in education (except medical or safety uses), biometric categorisation that infers protected attributes, and real-time remote biometric identification in publicly accessible spaces for law enforcement, with the exceptions in that point. Each point has its own limits. Two further prohibitions apply from 2 December 2026, covered in the next section.

  • High risk - Two paths. Annex III (Article 6(2)) lists use cases. Examples include biometrics, critical infrastructure, education and employment, credit scoring, law enforcement, migration, and the administration of justice. The list is longer than those examples. Annex I (Article 6(1)) covers an AI system that is a safety component of a regulated product, or is itself that product. These carry the heaviest requirements.

  • Limited risk - Systems that interact directly with people, like chatbots and AI-generated content. The main obligation here is transparency: users need to know they’re interacting with AI.

  • Minimal risk - Everything else. Spam filters, AI-powered search, recommendation engines, most internal tools. This tier has no conformity assessment and no Article 50 transparency duty. Article 4 still requires providers and deployers to take measures to support AI literacy, and it does not require any person to reach a set level. Voluntary codes of conduct are also encouraged.

The honest assessment? If you’re a typical tech company, most of your AI systems fall under minimal or limited risk. That’s not a reason to ignore the regulation - it’s a reason not to panic.

What is banned from 2 December 2026?

From 2 December 2026 the new bans are non-consensual intimate material and child sexual abuse material. Regulation (EU) 2026/1744 inserts them into Article 5(1) of the AI Act as points (ba) and (bb). Article 113 says they, as limited by Article 5(1a) and (1b), apply from 2 December 2026. The earlier prohibitions are untouched and have applied since 2 February 2025.

Point (ba) prohibits placing on the market, putting into service, or using an AI system that “generates or manipulates realistic images, videos, audio or similar material of an identifiable natural person’s intimate parts, or of an identifiable natural person engaged in sexually explicit activities,” without that person’s “freely-given, specific, informed, unambiguous and explicit consent” for that generation or manipulation.

Point (bb) prohibits placing on the market, putting into service, or using an AI system that generates or manipulates “material or performance within the meaning of Article 2, points (c) and (e), of Directive 2011/93/EU,” except where a “without right” defence applies under national law. Those points define child pornography and pornographic performance. Recital 11 describes that prohibition as child sexual abuse material, including wholly or partially synthetic material. Recital 10 is the reason for amending Article 5 for both new bans. The operative ban is the Directive’s definition, not the recital label.

Article 5(1a) narrows who the ban hits. For a provider, placing the system on the market or putting it into service is prohibited only in two cases. The first is where that generation or manipulation is the intended purpose. The second is where the system’s design, training, architecture, capabilities, or user-facing functionalities make it a reasonably foreseeable and reproducible outcome without significant technical modification, and the system does not have reasonable and adequate technical safety measures and other safeguards to reliably prevent that outcome, taking into account reasonably foreseeable misuse, and to correct observed or reported misuse.

For a deployer, use is prohibited only where the deployer uses the system for that purpose. Article 5(1b) says manipulation that does not increase the exposure of any depicted intimate parts, and does not alter the nature of any depicted sexually explicit activities, is not “manipulation” for point (ba).

If you ship a generative image, video, or audio system into the EU, this is a product question, not a policy poster. An intended purpose of that kind is prohibited. So is a reasonably foreseeable and reproducible outcome of that kind, without significant technical modification, when the system lacks reasonable and adequate safeguards to reliably prevent it and to correct observed or reported misuse. Using someone else’s model for that purpose is prohibited for the deployer. That’s the regulation’s line. It isn’t a design guide, and it isn’t legal advice.

What does high-risk actually require?

If any of your AI systems do land in the high-risk category, the requirements are substantial. You’ll need:

  • Risk management system - A documented, ongoing process for identifying and mitigating risks throughout the AI system’s lifecycle. Not a one-time assessment - a living system.

  • Data governance - Training, validation, and testing datasets must meet quality criteria. You need to address biases and ensure data is representative and relevant.

  • Technical documentation - Detailed records of how the system works, what it was designed to do, and how it was tested. Think of it as a comprehensive system dossier.

  • Human oversight - Mechanisms that allow humans to understand, monitor, and intervene in the AI system’s operation. The level of oversight should match the level of risk.

  • Accuracy, robustness, and cybersecurity - The system must perform consistently and be resilient against errors, faults, and adversarial attacks.

If this sounds familiar, it should. These requirements overlap significantly with ISO 42001 - the international standard for AI management systems. Organizations that pursue ISO 42001 certification are already building much of that infrastructure. Certification does not guarantee compliance.

What is the EU AI Act timeline?

The EU AI Act entered into force in August 2024. Enforcement is phased, and Regulation (EU) 2026/1744 replaced the original high-risk dates:

  • 2 February 2025 - Chapters I and II apply from this date, except Article 5(1)(ba), (bb), (1a) and (1b). The original prohibited practices are already in effect.
  • 2 August 2025 - General-purpose AI (GPAI) model obligations apply (already in effect). Regulation (EU) 2026/1744 did not move this date.
  • 2 August 2026 - General application, including Article 50 transparency. Recital 40 of Regulation (EU) 2026/1744 keeps this as the general date. It was not postponed with the high-risk rules.
  • 2 December 2026, new bans - Article 5(1), points (ba) and (bb), apply, as limited by Article 5(1a) and (1b). Point (ba) is non-consensual intimate material. Point (bb) is what recital 11 of Regulation (EU) 2026/1744 calls child sexual abuse material. The operative text is child pornography and pornographic performance, as Directive 2011/93/EU Article 2(c) and (e) define them.
  • 2 December 2026, marking - This date ends the Article 50(2) transitional period. Providers of AI systems, including general-purpose AI systems, that generate synthetic audio, image, video or text and were placed on the market before 2 August 2026 must comply with that marking duty by 2 December 2026 (Article 111(4)). A system placed on the market on or after 2 August 2026 does not get that window. Recital 38 of that regulation calls it four months.
  • 2 December 2027 - Chapter III, Sections 1, 2 and 3, except Article 6(5), for high-risk AI systems in Annex III (Article 6(2)). Examples include biometrics, employment, and credit scoring. Article 6(5) is not postponed with those sections.
  • 2 August 2028 - The same Chapter III sections, except Article 6(5), for high-risk AI systems in Annex I (Article 6(1)): AI that is a safety component of a regulated product, or is itself that product.
  • Systems already placed on the market - Article 111(2) applies those Chapter III duties to a high-risk system placed on the market or put into service before the relevant date only if the design changes significantly after it. Recital 39 of Regulation (EU) 2026/1744 lets further units of the same type and model continue while that design is unchanged. Systems intended for public authorities have until 2 August 2030.

December 2026 is the date to plan for if you place a generative system on the EU market. December 2027 is the near date if a system is high-risk under Annex III. Building a risk management system, documenting the system properly, and putting data governance in place takes months. If you haven’t started, start with the inventory.

What are the penalties?

The regulation has teeth:

  • Up to 35 million EUR or 7% of total worldwide annual turnover for the preceding financial year, for violations involving prohibited AI practices
  • Up to 15 million EUR or 3% of that turnover for other non-compliance
  • Up to 7.5 million EUR or 1% of that turnover for incorrect, incomplete, or misleading information given to notified bodies or national competent authorities in reply to a request

For an undertaking, each ceiling is the higher of the fixed amount and the percentage. Article 99(6) uses the lower of the two for SMEs, including start-ups. Article 99(6a) uses the lower of the two for small mid-cap enterprises on the 15 million EUR or 3% tier and the 7.5 million EUR or 1% tier. The prohibited-practice ceiling stays the higher of the two for a small mid-cap.

For context, GDPR’s maximum is 20 million EUR or 4% of turnover. The EU AI Act goes further. And if GDPR enforcement taught us anything, it’s that the EU means business.

When do the sandboxes have to be running?

Article 57(1), as amended by Regulation (EU) 2026/1744, says each Member State must have at least one national AI regulatory sandbox operational by 2 August 2027. That is August, not December. A Member State can build the sandbox with other Member States. The Commission can offer technical support.

The AI Office may establish a Union-level sandbox for systems covered by Article 75(1). If it does, that sandbox gives priority access to SMEs, including start-ups, and to small mid-caps. It does not move the national deadline, and it does not take over sandboxes that stay under national supervision.

A sandbox is a controlled place to develop, train, test, and validate a system for a limited time, under a plan agreed with the authority. It is not a waiver of Article 5, and it is not a certificate.

What did the Omnibus simplify for smaller companies?

Recital 6 of Regulation (EU) 2026/1744 says measures that the AI Act already gave to smaller operators should, where appropriate, also reach small mid-cap enterprises, without lowering the Act’s protection. One concrete piece is registration. A provider that keeps a system out of the high-risk tier under Article 6(3) still registers it in the EU database. The Omnibus simplified the Annex VIII fields for that registration. Recital 22 says the registration stays, because market surveillance and public accountability still need it. The provider still documents the Article 6(3) assessment before the system is placed on the market or put into service.

The fine ceilings in Article 99(6) and (6a) are the other piece people mix up. SMEs, including start-ups, are capped at the lower of the fixed amount and the percentage. Small mid-caps get that lower figure on the 15 million EUR or 3% tier and the 7.5 million EUR or 1% tier. The prohibited-practice ceiling stays the higher of the two for a small mid-cap.

Which Omnibus is actually in force?

Regulation (EU) 2026/1744 is in force. It has been since 27 July 2026. It is the Digital Omnibus on AI, and it is the instrument behind the dates on this page.

A different proposal would simplify the GDPR, ePrivacy, parts of the data acquis, and cyber incident reporting. On 21 October 2026 the Council agreed its negotiating position on that digital rulebook. A Council position is not a regulation. Do not rebuild a privacy program on it, and do not treat it as a reason to delay work the AI Act already requires.

Three notes sit next to this timeline, and none of them moves a date:

  • EN 18286 is the first harmonised standard cited for the Act. The citation supports a presumption for the Article 17 quality-management requirements the standard covers. It does not replace ISO 42001.
  • ISO/IEC 27090 is security guidance, published on 14 October 2026. It is not a certificate, and it is not in that harmonised-standard decision.
  • AI literacy is Article 4. It has applied since 2 February 2025. It asks providers and deployers to support people’s skills. It does not set a pass mark.

What should you do right now?

Here’s a practical starting point - three steps that apply regardless of your company’s size:

  1. Build an AI system inventory - Map every AI system in your organization, including third-party tools and APIs. You’d be surprised how many AI systems fly under the radar. That chatbot your marketing team set up? The AI feature in your CRM? They count.

  2. Classify your risk levels - For each system, determine which risk category it falls into under the EU AI Act. Be honest but don’t over-classify. Most internal tools and customer-facing chatbots are limited or minimal risk.

  3. Identify gaps and prioritize - For any high-risk systems, compare your current practices against the regulation’s requirements. Focus on the gaps that will take the longest to close: risk management systems, documentation, and data governance processes.

This isn’t about checking boxes. An AI system inventory and risk classification done right becomes a strategic asset - it gives you visibility into how AI flows through your organization and where the real risks (and opportunities) live.

How does ISO 42001 fit in?

ISO 42001 and the EU AI Act aren’t the same thing, but they’re deeply complementary. ISO 42001 provides the management system framework - the policies, processes, and controls - that organizations need to govern AI responsibly. The EU AI Act defines the legal requirements.

If you build an AI management system aligned with ISO 42001, you’re already covering significant ground on EU AI Act compliance: risk management, data governance, documentation, human oversight, and continuous improvement. Pursuing certification doesn’t guarantee compliance, but it demonstrates a systematic approach that regulators respect.

We’ve seen this pattern before with ISO 27001 and GDPR. Organizations with mature information security management systems had a much easier time meeting GDPR requirements. The same dynamic applies here.

How we can help

We’ve spent a decade helping organizations navigate EU regulatory frameworks - from GDPR and ISO 27001 to the newer world of AI governance. The EU AI Act is complex, but compliance doesn’t have to be overwhelming.

We’ll help you build an AI system inventory, classify your risk levels honestly, and create a compliance roadmap that aligns with the regulation’s phased timeline. If you’re also considering ISO 42001 certification, we’ll design a unified approach that covers both without duplicating effort.

And if after the assessment your honest answer is “most of our systems are minimal risk” - we’ll tell you that too. No scope inflation, no manufactured urgency.

Let’s talk - we’ll help you figure out exactly where your AI systems stand.