EU AI Act Timeline After the Omnibus: What Product Teams Must Do

Kate Kubasova
November 17, 2026 · 9 min read

The EU AI Act timeline changed in July 2026, when the Digital Omnibus on AI amended the regulation. Most rules have applied since 2 August 2026, including the Article 50 transparency rules for chatbots and AI-generated content. Generative AI systems that were already on the market before that date have until 2 December 2026 to add machine-readable marking. The high-risk rules moved to 2 December 2027 for Annex III uses such as recruitment and credit scoring, and to 2 August 2028 for AI built into regulated products.

This guide is for teams that ship AI features (assistants, agents, RAG search, generated images or text) to users in the EU. It covers the dates, what the Omnibus changed, what Article 50 asks of your product today, how to check whether a feature is high-risk, and what to do in the next twelve months. It is general information, not legal advice.

EU AI Act timeline at a glance

Date What applies
1 August 2024 The AI Act enters into force
2 February 2025 Prohibited AI practices and the AI literacy duty (Article 4)
2 August 2025 Obligations for general-purpose AI (GPAI) model providers; governance and penalties
27 July 2026 The Digital Omnibus on AI enters into force
2 August 2026 Most remaining rules apply and enforcement starts, including all of Article 50 for new systems
2 December 2026 Marking deadline for generative AI systems placed on the market before 2 August 2026; new bans on generating non-consensual intimate imagery and child sexual abuse material
2 August 2027 Each Member State must have at least one AI regulatory sandbox
2 December 2027 High-risk rules for Annex III systems (employment, credit, education, biometrics and others)
2 August 2028 High-risk rules for AI in products covered by EU product law (Annex I)

Sources: AI Act Service Desk implementation timeline and European Commission, July 2026.

What the Digital Omnibus changed, and what it didn't

The Commission proposed the Omnibus in November 2025. Parliament and Council reached a provisional agreement on 7 May 2026. The text was published as Regulation (EU) 2026/1744 on 24 July 2026 and entered into force on 27 July 2026 (European Parliament Legislative Train).

What moved:

  • High-risk deadlines. Annex III moved from August 2026 to 2 December 2027, and Annex I moved from August 2027 to 2 August 2028. The original proposal tied these dates to harmonised standards being ready; the final law sets fixed dates instead.
  • AI literacy (Article 4) was rewritten. Providers and deployers must still take measures to support AI literacy among staff who work with AI. The text now says this doesn't require them to guarantee a specific level of literacy (Article 4).
  • Small mid-caps get SME-style relief. That includes simplified technical documentation, proportionate quality management and priority sandbox access. Orrick describes a small mid-cap as a company that isn't an SME but has fewer than 750 staff and turnover of €150 million or less, or a balance sheet of €129 million or less (Orrick, July 2026).
  • Registration got lighter but stayed. If you decide an Annex III system isn't high-risk under Article 6(3), you must still register it, but with less information.

What didn't move: Article 50. Transparency duties for chatbots, AI agents and AI-generated content have applied since 2 August 2026. Only systems already on the market before that date get extra time, until 2 December 2026, and only for the marking duty. Some articles written before the final text said marking moved to December for everyone; the final text limits it to existing systems.

The EU AI Act timeline for product teams
2 Feb2025
Prohibited practices and AI literacy
2 Aug2025
General-purpose AI model rules
Obligations for model providers, governance, penalties
2 Aug2026
Article 50 transparency applies
AI disclosure in chats and agents; marking of AI-generated content for new systems
Nov 2026
2 Dec2027
High-risk rules, Annex III
Employment, credit, education, biometrics (was August 2026)
2 Aug2028
High-risk rules, Annex I
AI built into products under EU product law (was August 2027)
Sources: EU AI Act Service Desk implementation timeline; European Commission, "AI Omnibus enters into force" (July 2026). General information, not legal advice.

Does the EU AI Act apply to your company?

Location doesn't take you out of scope. The Act covers providers that place AI systems on the EU market wherever they are established. It also covers providers and deployers outside the EU when the system's output is used in the EU (Article 2). A startup in Israel, the UK or the US that sells an AI feature to EU customers is a provider in scope.

Two roles matter for most product teams:

  • Provider: the company that develops an AI system, or has it developed, and offers it under its own name. If your SaaS product includes an LLM assistant, you are usually the provider of that feature, even if the model behind it comes from OpenAI, Anthropic or Google.
  • Deployer: the organisation that uses the system under its own authority. Your business customers are usually deployers.

Building on someone else's model doesn't move responsibility upstream. Under Article 25, a company that puts its name on a system, substantially modifies it or repurposes it for a high-risk use becomes its provider (Article 25).

Article 50: what applies to chatbots, agents and AI content now

Article 50 has four parts. The first two apply to providers (most product teams), and the last two to deployers (Article 50).

1. Tell users they are talking to AI (providers)

If people interact with your system, it must tell them they are dealing with AI, unless that is obvious to a reasonably well-informed person. The Commission's July 2026 guidelines say the disclosure must come at the first interaction. A line in the terms and conditions or the documentation is not enough (Article 50 guidelines, C(2026) 5054).

For AI agents that email, call or message people on a user's behalf, Faegre Drinker reads the guidelines as requiring the agent to say whom it acts for. Machine-to-machine steps such as API calls need no disclosure (Faegre Drinker, July 2026).

2. Mark AI-generated content (providers)

Synthetic audio, images, video and text must carry machine-readable marking and be detectable as AI-generated, as far as that is technically feasible. Standard editing assistance and outputs that don't substantially change the user's input are exempt. In practice teams combine signed metadata with watermarking.

The Commission published the final Code of Practice on Transparency of AI-Generated Content on 10 June 2026 (European Commission). Signing is voluntary, and the Commission says signing is not conclusive evidence of compliance. More than 180 organisations had signed by 31 July 2026.

3. Emotion recognition and biometric categorisation (deployers)

Deployers must tell people when they are exposed to these systems.

4. Deepfakes and public-interest text (deployers)

Deployers must disclose deepfakes, with a lighter form for clearly artistic, satirical or fictional work. AI-generated text published to inform the public on matters of public interest must also be disclosed, unless a human has reviewed it and someone holds editorial responsibility. That category is narrower than "any text an AI wrote". Check with counsel how it applies to your use case.

Four transparency duties, in force since 2 August 2026
Providers
The company that builds the AI feature and offers it under its own name
Art. 50(1)
Tell users they're talking to AI
At the first interaction. A line in the terms isn't enough.
Art. 50(2)
Mark AI-generated content
Machine-readable marking on audio, images, video and text. Existing systems: by 2 December 2026.
Deployers
The organisation that uses the AI system under its own authority
Art. 50(3)
Disclose emotion recognition
Also biometric categorisation, to the people exposed to it.
Art. 50(4)
Label deepfakes and public-interest text
Unless a human reviewed the text and holds editorial responsibility.
€15m or 3%
Maximum fine for breaching Article 50, whichever is higher. SMEs and start-ups face whichever is lower.
Sources: AI Act Articles 50 and 99 (AI Act Service Desk); Commission guidelines on Article 50, C(2026) 5054 (July 2026). General information, not legal advice.

Is your AI feature high-risk? Annex III areas for software companies

High-risk rules don't apply until 2 December 2027, but a product built now will still be on sale then. Check your features against the Annex III areas most relevant to software companies (Annex III):

  • Employment: targeting job ads, filtering applications, evaluating candidates, decisions on promotion or termination, allocating tasks based on personal traits, and monitoring worker performance. Many HR-tech AI features fall here.
  • Credit and insurance: creditworthiness assessment and credit scoring (fraud detection is excluded), and risk assessment and pricing for life and health insurance.
  • Education: admissions, evaluating learning outcomes, assigning education levels and proctoring.
  • Essential services: eligibility for public benefits, and emergency-call triage.
  • Biometrics: remote identification, categorisation by sensitive traits and emotion recognition.

There is a way out under Article 6(3). A system in an Annex III area is not high-risk if it only performs a narrow procedural task, improves the result of work a human already did, or does preparatory or follow-up work. But any system that profiles people is always high-risk. If you rely on the exemption, document the assessment and register the system (Article 6).

One more planning point: a type of high-risk system lawfully placed on the market before its deadline can continue to be sold as long as its design doesn't change significantly (Cooley, August 2026). Any substantial redesign after the deadline brings the system fully under the rules, so this helps little for products that change often.

EU AI Act fines

Breach Maximum fine
Prohibited AI practices €35 million or 7% of worldwide annual turnover
Most other obligations, including Article 50 €15 million or 3%
Misleading information to authorities €7.5 million or 1%

For most companies the higher of the two amounts applies. For SMEs and start-ups the lower amount applies. Small mid-caps get the lower amount for the second and third tiers only (Article 99). Some articles state the transparency tier as €7.5 million or 1.5%. Article 99 sets it at €15 million or 3%.

What to do now: an EU AI Act checklist for product teams

The work falls into two windows. Three tasks come before 2 December 2026, when the marking deadline for existing systems arrives. Four belong to 2027, before the high-risk rules apply on 2 December 2027.

Most of it is engineering work: disclosure UX, marking pipelines, logging and documentation. Product, engineering and legal should own the list together.

Before 2 December 2026Now
List every AI feature you provide or use
Include features built by contractors and third-party models inside your product.
Disclose AI at the first interaction
Check every chat and agent interface, in every language and accessible view you support. Keep screenshots or other evidence.
Add machine-readable marking to generated content
Required if your product generates images, audio, video or text and was on the market before 2 August 2026. Decide whether to sign the Code of Practice.
During 2027High-risk deadline: 2 Dec 2027
Classify each feature against Annex III
Document any Article 6(3) exemption. Pay close attention to HR, credit, insurance and education features; profiling is always high-risk.
Ask your model providers for their downstream documentation
GPAI model providers must give it to companies building on their models (Article 53). For high-risk uses, put supplier terms in writing.
Plan the engineering work for any high-risk feature
Risk management, data governance, logging, human oversight and technical documentation, in place by 2 December 2027.
Keep AI literacy measures going
For everyone who builds or operates AI features. The duty has applied since February 2025.
General information, not legal advice. Check your own case with counsel.

Most of this checklist is engineering work: disclosure UX, metadata and watermarking pipelines, logging and evaluation. Our guide on what to have in place before you ship an AI feature covers the technical side.

Summary

Article 50 has applied since 2 August 2026. Every AI chat or agent must disclose itself at first contact, and generated content must carry machine-readable marking. Systems already on the market before that date have until 2 December 2026 for marking. The Omnibus moved the high-risk rules to 2 December 2027 for Annex III and 2 August 2028 for Annex I, so HR-tech, fintech and edtech teams have about a year to classify their features and build the controls. Non-EU companies selling into the EU are in scope.

If you need engineers to implement disclosure, marking, logging or evaluation in your product, 5Blue Software helps teams with AI-powered software development and AI implementation. Talk to us about your roadmap.

FAQ

Has the EU AI Act been delayed?

Partly. The Digital Omnibus, in force since 27 July 2026, moved the high-risk deadlines to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). The transparency rules in Article 50 were not delayed and have applied since 2 August 2026.

Do chatbots have to say they are AI in the EU?

Yes, unless it is obvious to a reasonably well-informed user. The disclosure must come at the first interaction, and a mention in the terms and conditions is not enough, according to the Commission's Article 50 guidelines.

Is there a grace period for existing chatbots?

Not for the disclosure duty. The only extra time is for machine-readable marking of generated content: generative systems placed on the market before 2 August 2026 have until 2 December 2026.

Does the EU AI Act apply to companies outside the EU?

Yes. It applies to providers that place AI systems on the EU market and to providers and deployers whose AI output is used in the EU, wherever they are based.

What is the fine for breaking Article 50?

Up to €15 million or 3% of worldwide annual turnover, whichever is higher. SMEs and start-ups face whichever amount is lower.

Is an AI recruiting feature high-risk?

Usually. Filtering applications, evaluating candidates and monitoring worker performance are listed in Annex III, so those rules apply from 2 December 2027. A feature that only does a narrow procedural task may be exempt under Article 6(3), but any feature that profiles people is high-risk.

Share Post

More articles