Last updated: 6 November 2026. We update this page when AI Act dates or guidance change.
The EU AI Act has now been amended once. Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, was signed on 8 July 2026, published in the Official Journal on 24 July 2026 and entered into force on 27 July 2026. It pushes back the high-risk rules, adds two new bans, softens the AI literacy duty, extends relief to smaller companies and gives the EU's AI Office more direct supervision.
The basic structure of the Act is unchanged: the risk-based tiers, the list of high-risk uses, the obligations for general-purpose AI models and the transparency rules all stay. This page covers only what the amendment changed and when each part applies. For an introduction to the Act itself, see our EU AI Act guide. You can read the official text of Regulation (EU) 2026/1744 on EUR-Lex.
The AI Act timeline after the amendments
| Date | What applies | Changed by 2026/1744? |
|---|---|---|
| 2 February 2025 | Bans on prohibited AI practices. AI literacy duty | AI literacy reworded from 27 July 2026 |
| 2 August 2025 | Obligations for general-purpose AI models. Governance and national penalty rules | No |
| 2 August 2026 | Most of the Act, including the Article 50 transparency rules. Commission fines for general-purpose AI providers | No, apart from the watermarking grace period below |
| 2 December 2026 | Two new prohibited practices. Machine-readable marking of AI-generated content for systems already on the market before 2 August 2026 | New |
| 2 August 2027 | General-purpose AI models placed on the market before 2 August 2025 must comply. Each EU country must have an AI regulatory sandbox running | Sandbox deadline moved from 2 August 2026 |
| 2 December 2027 | High-risk rules for Annex III systems (for example AI used in hiring, credit scoring, education and essential services) | Moved from 2 August 2026 |
| 2 August 2028 | High-risk rules for AI in regulated products under Annex I (for example machinery, medical devices and toys) | Moved from 2 August 2027 |
1. High-risk obligations pushed back
This was the main reason for the amendment. The original Act would have applied the full high-risk regime from 2 August 2026, but the harmonised standards and guidance that companies need to comply were not ready. The EU fast-tracked the AI part of the omnibus so the new dates would be law before the old deadline arrived.
-
Annex III systems, the standalone high-risk uses such as recruitment tools, creditworthiness checks, exam scoring and access to public services, now face the full requirements from 2 December 2027.
-
Annex I systems, AI that is a safety component of a product already covered by EU product safety law, follow from 2 August 2028.
The Commission had proposed tying the start date to when standards became available. The final text sets fixed dates instead, which gives companies a firm deadline to plan against. The requirements themselves have not changed: risk management, data governance, technical documentation, logging, human oversight, accuracy and a quality management system are all still required. The deadline moved, not the obligations.
2. Two new prohibited practices
From 2 December 2026, Article 5 bans two more uses of AI:
-
using AI systems to generate or manipulate child sexual abuse material, and
-
using AI systems to generate non-consensual intimate or sexually explicit images, video or audio of identifiable people, often called "nudifier" apps.
Providers of general-purpose image, video and audio generators need to assess this kind of foreseeable misuse when they design and deploy their tools. Breaches of Article 5 carry the Act's highest fines: up to €35 million or 7% of worldwide annual turnover.
3. A softer AI literacy duty
Article 4 has applied since 2 February 2025. The original text required providers and deployers to take measures to ensure, to their best extent, a sufficient level of AI literacy among staff and others using AI on their behalf. The amended Article 4 now requires them to take measures to support the development of AI literacy, and adds that this does not require them to guarantee any specific level of literacy for any individual.
The duty still applies to every organisation that provides or uses AI systems, whatever the risk level. In practice, the change makes it a duty to make a documented effort rather than to achieve a result. Training, internal guidance and clear rules on which AI tools staff may use are still the way to meet it.
4. More relief for SMEs and small mid-caps
The original Act already gave small and medium-sized enterprises some relief. The amendment extends it to small mid-cap enterprises as defined in Commission Recommendation (EU) 2025/1099: companies with fewer than 750 employees and either annual turnover of up to €150 million or a balance sheet of up to €129 million. The relief covers:
-
a simplified technical documentation form that notified bodies must accept,
-
quality management obligations that can be proportionate to the company's size, and
-
lower fine caps.
The GDPR record-keeping deal in the separate Omnibus IV package uses a wider small mid-cap definition (fewer than 1,000 employees). The two definitions are not the same, so check each one separately. Our guide to GDPR simplification for small businesses covers the GDPR side.
5. Registration kept, with less detail
Under Article 6(3), a provider can decide that a system listed in Annex III is not actually high-risk, for example because it only performs a narrow preparatory task. The Commission proposed dropping the requirement to register these systems in the EU database. The final text keeps registration but reduces the information that has to be submitted. If you rely on Article 6(3), you still need to document your assessment and register the system.
6. A wider basis for using sensitive data to detect bias
The original Act let providers of high-risk systems process special category personal data, such as ethnicity or health data, where strictly necessary to detect and correct bias (Article 10(5)). The amendment deletes that paragraph and replaces it with a new, general Article 4a. It also covers deployers of high-risk systems and providers and deployers of other AI systems and models.
The safeguards are cumulative. The processing must be strictly necessary, other data such as synthetic or anonymised data must not be enough, and the data has to be protected and deleted once the bias is corrected. Article 4a is a permission, not a duty: it does not oblige anyone to run bias tests. The rest of the GDPR still applies, so you still need to be transparent about the processing and carry out a DPIA where one is required.
7. More supervision by the AI Office
The Commission's AI Office already supervised general-purpose AI models. It now also has exclusive competence, with new investigation and fining powers, over:
-
AI systems built on a general-purpose model where the model and the system come from the same provider or corporate group, and
-
AI systems that are, or are part of, a very large online platform or search engine under the Digital Services Act.
There are exceptions for regulated products, critical infrastructure, law enforcement, border management, financial institutions and the courts. A company that builds a product on a third party's model stays under its national market surveillance authority.
What did not change
-
The structure of the Act: prohibited practices, high-risk systems, transparency obligations and minimal-risk AI.
-
General-purpose AI model obligations, which have applied since 2 August 2025.
-
Article 50 transparency rules, which have applied since 2 August 2026. People must be told when they are dealing with a chatbot, and deepfakes must be disclosed. The only relief is the four-month marking grace period for generative systems already on the market.
-
The fine levels: up to €35 million or 7% of turnover for prohibited practices, €15 million or 3% for most other breaches, and €7.5 million or 1% for supplying incorrect information.
-
The territorial scope: non-EU companies are still covered if their AI systems are used in the EU.
How this fits with the GDPR changes
The Digital Omnibus on AI was split from the wider Digital Omnibus package, which also proposes GDPR changes, such as confirming that legitimate interest can support AI development. That data part is still being negotiated: the Council agreed its position on 21 October 2026 and the European Parliament has not voted yet. Until it is adopted, training or running AI on personal data is governed by the current GDPR, including the rules on automated decision-making. Our overview of GDPR changes and updates tracks both files.
What to do now
-
Check your Article 50 obligations today. Chatbot disclosures and deepfake labelling already apply. If you provide a generative AI system that was on the market before 2 August 2026, machine-readable marking of its output is due by 2 December 2026.
-
Review generative tools against the new bans before 2 December 2026, especially image, video and voice features that could be misused to create intimate content of real people.
-
Document your AI literacy measures. Keep a record of training, guidance and approved tools. The duty is softer, but it has not gone away.
-
Use the extra time for high-risk systems. Build an inventory of your AI systems, classify each one and start on risk management and technical documentation now. December 2027 is close for a full compliance programme.
-
Check your size category against the small mid-cap thresholds to see whether the simplified documentation and lower fine caps apply to you.
-
Plan bias testing around Article 4a and the GDPR together. Record why the sensitive data is strictly necessary and how it will be deleted. Our guide to AI training data governance covers the data side.
ComplyDog helps you keep the GDPR side of your AI use in order, from records of processing to DPIAs, as the AI Act deadlines approach.
FAQ
Has the EU AI Act been delayed?
Partly. The high-risk obligations were pushed back to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). The bans, the AI literacy duty, the general-purpose AI rules and the Article 50 transparency rules were not delayed, apart from a short grace period for watermarking.
Is the AI Act still in force?
Yes. It has been in force since 1 August 2024, and most of it has applied since 2 August 2026. Regulation (EU) 2026/1744 amends it but does not replace it.
What is the Digital Omnibus on AI?
It is the name for Regulation (EU) 2026/1744, the law that amended the AI Act in July 2026. The Commission proposed it on 19 November 2025 alongside a separate Digital Omnibus that amends the GDPR and other data laws. That separate proposal has not been adopted.
Do I still need AI literacy training?
You still need to take measures to support AI literacy among the people who use AI on your behalf. The amendment removed the requirement to ensure a particular level, but training and guidance remain the simplest way to show you have met the duty.
Does the AI Act amendment change the GDPR?
Not directly. The new Article 4a gives a legal basis for using sensitive data to detect and correct bias, but every other GDPR obligation still applies. Changes to the GDPR itself are in the separate Digital Omnibus, which is still being negotiated.