What Applies Under the EU AI Act From August 2026

The High-Risk Deadline Moved but 2 August Still Arrived

For two years the date on every AI compliance roadmap was 2 August 2026. Six days before it arrived, the Digital Omnibus on AI entered into force and pushed the heaviest high-risk obligations out to December 2027 and August 2028. Many teams read that as a pause. It was not. If you are asking what applies under the EU AI Act from August 2026, the answer is most of the Act, just not the part you budgeted for.

The reason is a mismatch between where the work was planned and where the obligations landed. The 2024 text made 2 August 2026 the start of both general application and the Annex III high-risk regime, so programmes were built around conformity assessments and technical files. That work is still needed. What changed is the order, because Article 50 transparency and the enforcement framework applied on schedule.

What is the Digital Omnibus on AI?

The Digital Omnibus on AI is Regulation (EU) 2026/1744, a targeted amendment to the EU AI Act signed on 8 July 2026. It defers the Chapter III high-risk obligations, adds two prohibited practices, softens the AI literacy duty and strengthens the AI Office. It does not repeal or pause the Act.

It was published in the Official Journal of the EU on 24 July 2026, so guidance that still calls the deferral provisional is out of date.

The deferral covers Sections 1 to 3 of Chapter III only. For stand-alone Annex III systems, such as AI used in recruitment, credit scoring, education or biometrics, those rules now apply from 2 December 2027. For AI that is a safety component of an Annex I product, such as a medical device, they apply from 2 August 2028 instead of 2 August 2027. Both dates are fixed.

Take a Swedish retailer using a vendor tool to screen job applicants and a chatbot for customer queries. The chatbot’s disclosure duty has applied since 2 August 2026. The screening tool is an Annex III employment use, so its high-risk duties start on 2 December 2027.

The EU AI Act Implementation Timeline After the Omnibus

2 February 2025	Original Article 5 prohibitions and the Article 4 AI literacy duty apply.
2 August 2025	Rules for general-purpose AI models, governance and penalties apply.
27 July 2026	The Digital Omnibus enters into force and the softer Article 4 takes effect.
2 August 2026	General date of application. Article 50 transparency duties apply.
2 December 2026	New nudifier and CSAM bans apply. Generative systems already on the market must carry machine-readable marking.
2 December 2027	High-risk obligations apply to stand-alone Annex III systems.
2 August 2028	High-risk obligations apply to AI embedded in Annex I products.
2 August 2030	Deadline for high-risk systems used by public authorities.

Am I a Provider or a Deployer?

The AI Act assigns duties mainly by role. Under Article 3 a provider develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. A deployer uses an AI system under its own authority in a professional context. Most organisations are both.

The line can move. Under Article 25 a deployer becomes the provider of a high-risk system if it puts its own name or trademark on it, makes a substantial modification to it or changes its intended purpose so that it becomes high-risk. Fine-tuning a vendor model and releasing it under your own brand is one way that happens unnoticed.

Who Carries Which Duty Today

• Providers : Tell people they are interacting with AI unless that is obvious, and mark generative outputs in a machine-readable format.
• Deployers : Label deepfakes and AI-generated public-interest text, and inform people exposed to emotion recognition or biometric categorisation.
• Both Roles : Take measures to support AI literacy and avoid every prohibited practice in Article 5.

What to Do Now, Before December 2026 and Before December 2027

Start with an inventory. List every AI system you build, buy or embed, record your role for each and flag anything that talks to people, generates content or falls under Annex III.

Then check the deepfake labelling requirement in Article 50(4), which is narrower than many assume. It covers image, audio or video that resembles real people, places or events and would falsely appear authentic. It also covers AI-generated text published to inform the public on matters of public interest, unless a person has reviewed it and taken editorial responsibility.

Before 2 December 2026, providers of generative systems already on the market before 2 August 2026 must add machine-readable marking. The same date brings new Article 5 bans on AI that generates non-consensual intimate imagery of identifiable people or child sexual abuse material. Providers are caught where that output is the purpose or a foreseeable result without adequate safeguards, and deployers only if they use a system for that purpose.

Before 2 December 2027, use the extra time for Annex III work. Classify each high-risk use case, collect provider documentation and plan a fundamental rights impact assessment wherever Article 27 requires one. That assessment can now cross-refer to an existing GDPR impact assessment.

Grandfathering and Significant Changes in Design

Article 111(2) protects high-risk systems placed on the market before their application date. They fall under the high-risk rules only if their design changes significantly afterwards, and the Omnibus confirmed that further units of an unchanged type and model share that protection. Systems intended for use by public authorities must still comply by 2 August 2030.

Keep the terms apart. Grandfathering turns on a significant change in design, while the Article 25 role transfer turns on a substantial modification. Date every model update so you can show which side of each line it falls on.

The EU AI Act for Swedish Companies

The AI Act applies directly in Sweden, but national enforcement is still being built. Under a Government decision of 4 June 2026, PTS (Post- och telestyrelsen), IMY, Finansinspektionen, Läkemedelsverket and Swedac act as national competent authorities until 31 December 2026. PTS is the single point of contact and supervision is divided by sector.

The inquiry report SOU 2025:101 proposes a permanent model with PTS as coordinating market surveillance authority. Until legislation follows, treat PTS as the first call when ownership is unclear.

The Risks Article 50 Was Written For

Transparency rules assume an honest operator and attackers do not label their deepfakes. The practical risk lies in impersonation, in unreliable output from systems you deploy and in attacks on the AI itself, which Article 15 names as data poisoning, model poisoning, adversarial examples and confidentiality attacks.

The Arup Deepfake Video Call

In early 2024 a finance employee at the Hong Kong office of engineering firm Arup joined a video call in which the chief financial officer and other colleagues were all deepfakes. According to Hong Kong police, the employee then made 15 transfers totalling HK$200 million, about US$25.6 million, to five local bank accounts. Arup confirmed the incident in May 2024.

The control that would have stopped it is out-of-band verification of every payment request through a separately known channel.

Moffatt v Air Canada

On 14 February 2024 British Columbia’s Civil Resolution Tribunal held Air Canada liable after its website chatbot wrongly told a customer that a bereavement fare could be claimed after travel. It rejected the argument that the chatbot was responsible for its own actions and ordered the airline to pay C$812.02 in damages, interest and fees.

Disclosing that users are talking to AI, as Article 50(1) requires, does not move that liability. The control is testing chatbot answers against your own published policies.

Penalties for Non-Compliance

Article 99 sets three fine tiers, each capped at a fixed amount or a share of worldwide annual turnover, whichever is higher. For SMEs the lower figure applies, and the Omnibus extends that to small mid-cap enterprises for the two lower tiers. Member States set national penalty rules, so Swedish exposure depends on the legislation proposed in SOU 2025:101.

The Fine Bands at a Glance

Article 5 prohibited practices carry fines of up to EUR 35 million or 7% of worldwide annual turnover. Most other obligations, including Article 50, carry up to EUR 15 million or 3%. Incorrect or misleading information to authorities carries up to EUR 7.5 million or 1%.

How to Maintain Compliance and Secure Your AI Systems

The AI security requirements in the EU AI Act sit mainly in Article 15. High-risk systems must reach and keep an appropriate level of accuracy, robustness and cybersecurity, including resilience against attempts to exploit vulnerabilities. For Annex III systems that applies from 2 December 2027, but the controls take longer to build than the paperwork.

Review the AI inventory whenever a model, vendor or use case changes. Red-team generative and agentic systems for prompt injection and data leakage, even though the Act does not name prompt injection. Log inputs, outputs and agent actions so decisions can be reconstructed. A high-risk system meeting the conditions in Article 12(1) of the Cyber Resilience Act is deemed to meet the Article 15 cybersecurity requirements.

Conclusion

The Digital Omnibus changed when the high-risk chapter applies, not whether it applies. Article 50 has been live since 2 August 2026 and the next hard date is 2 December 2026. Use it to finish your AI inventory, settle provider and deployer roles and test every system that talks to customers or generates content.

Frequently Asked Questions

What applies from 2 August 2026 under the EU AI Act?

It is the AI Act’s general date of application. Article 50 transparency duties apply, so chatbots must disclose that they are AI and deployers must label deepfakes. National supervision and AI Office enforcement also apply. Annex III high-risk obligations do not, because the Digital Omnibus moved them to 2 December 2027.

When is the EU AI Act high-risk deadline now?

For stand-alone high-risk systems listed in Annex III, such as AI used in hiring, credit scoring or education, the obligations apply from 2 December 2027. For AI embedded as a safety component in Annex I products, they apply from 2 August 2028. Both dates are fixed in Regulation (EU) 2026/1744.

If I buy an AI tool from a vendor and use it as-is, am I always just a deployer?

Usually, but not always. Using a system under your own authority makes you a deployer. Under Article 25 you become the provider of a high-risk system if you put your name or trademark on it, substantially modify it or change its intended purpose so that it becomes high-risk.

Who enforces the EU AI Act in Sweden?

Under a Government decision of 4 June 2026, PTS, IMY, Finansinspektionen, Läkemedelsverket and Swedac act as national competent authorities until 31 December 2026. PTS is the single point of contact and supervision is split by sector. The inquiry report SOU 2025:101 proposes a permanent structure.

What did the Digital Omnibus change in Article 4?

Article 4 no longer requires providers and deployers to ensure a sufficient level of AI literacy. Since 27 July 2026 they must take measures to support AI literacy among staff and others using AI on their behalf, without guaranteeing any specific level. The duty remains binding.

eBuilder Security’s AI Security / AIDR service covers the security side of an AI Act programme. See the AI Security service page to learn more.

This post is also available in: