The AI Act for companies: what applies now and what comes in 2027 and 2028
The AI Act's prohibitions, the obligations for general-purpose AI models and, since 2 August 2026, its general regime, including transparency, already apply. Obligations for high-risk systems arrive on 2 December 2027 (Annex III) and 2 August 2028 (Annex I), after the postponement set by Regulation (EU) 2026/1744.
By Rafael Tamames, founder of Vivid Vision
What is this page for?
For making decisions, not for studying the regulation. Here you will find the timeline and the decisions that belong to leadership. Risk levels, high-risk obligations and penalties are covered in detail in the guide The AI Act for companies .
The AI Act is Regulation (EU) 2024/1689. It has been in force since 1 August 2024 and applies in phases. Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, was published in the Official Journal of the EU on 24 July 2026, entered into force on 27 July and postponed the high-risk obligations.
Which parts of the AI Act already apply?
| Since | What applies | What it means for the company |
|---|---|---|
| 2 February 2025 | Ban on unacceptable-risk practices, such as social scoring or emotion recognition at work | Check that no system in use, in-house or third-party, falls into a prohibited practice |
| 2 August 2025 | Obligations for general-purpose AI models, such as large language models | They fall on the model providers; users need to know which models they have in use and for what |
| 2 August 2026 | General date of application of the regulation, including transparency | A chatbot or AI-generated content must be identified as such |
Applying now does not mean auditing everything tomorrow. It means knowing which AI systems the company has and the risk level of each one.
What changes on 2 December 2026?
It is a date added by the Omnibus and it is only weeks away. Two things start to apply that day.
- New prohibited practices
- AI systems that generate non-consensual intimate content or child sexual abuse material are banned.
- Marking synthetic content
- The deadline expires for marking content generated by systems placed on the market before 2 August 2026.
If your company offers customers AI-generated content, check how it is marked before that date.
What comes in 2027 and 2028?
| Date | Which systems | What it means for the company |
|---|---|---|
| 2 December 2027 | Annex III high risk, such as systems used in employment, education, credit or biometrics | If you use AI to recruit staff or score credit, you must use it according to the provider's instructions, keep human oversight, monitor how it works and keep the logs |
| 2 August 2028 | High risk embedded in regulated products under Annex I, such as machinery or medical devices | It mainly affects the manufacturers of those products and those who build them into their activity |
The deadline has moved, not the obligations. Preparing a high-risk system takes time: inventory, classification, owners, documentation and human oversight. That is why preparation starts now, not in 2027.
Which decisions should a CEO take now?
Five decisions a CEO cannot fully delegate and should settle before 2027.
- Who is accountable for AI
- Assign executive responsibility: a Chief AI Officer (CAIO) or a member of the leadership team, such as the CIO, CDO or COO. Each system in production also needs a business owner.
- Which AI we have
- Commission a single inventory of the AI systems in use, including third-party ones and those teams use on their own.
- What is high risk
- Classify that inventory by risk level and prioritise Annex III systems, whose deadline is 2 December 2027.
- What we ask of vendors
- Anyone using a high-risk system must follow the provider's instructions, so each vendor's documentation becomes part of the buying decision.
- How we prepare the teams
- Train executives and teams in AI applied to the business. This is the work of Activate .
More detail: How to govern AI in your company: a 4-step model · Guide: AI governance framework, roles and CAIO
Does it affect companies that only use third-party AI or are based outside the EU?
Yes, in both cases. The AI Act applies to those who develop or market AI systems (providers) and to those who use them in their professional activity (deployers) when the system is used in the European Union, even if the company is based elsewhere. A Latin American multinational with operations in Europe is covered through those operations.
In Spain, enforcement is overseen by AESIA , together with other authorities depending on the sector. Financial entities must also comply with DORA, fully applicable since 17 January 2025.
What happens if a company does not comply?
Article 99 sets fines of up to €35 million or 7% of worldwide annual turnover for using prohibited practices, up to €15 million or 3% for breaching other obligations, and up to €7.5 million or 1% for supplying incorrect information. The higher amount applies; for SMEs, the lower one.
More detail: Guide: the AI Act for companies, obligations and penalties
Frequently asked questions
Has the AI Act been delayed?
Only part of it. Regulation (EU) 2026/1744 postponed the obligations for high-risk systems to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). The prohibitions, the obligations for general-purpose models and the general regime already apply.
Which date should a CEO have in the diary?
2 December 2026, for the new prohibitions and the marking of synthetic content, and 2 December 2027 if the company uses AI in employment, education, credit or biometrics.
Does it affect companies that only use third-party AI?
Yes. Anyone who uses an AI system in their professional activity is a deployer. If the system is high-risk, they must use it according to the instructions, keep human oversight, monitor how it works and keep the logs.
Does the Omnibus change anything for SMEs?
It includes measures for SMEs and small mid-cap companies. For fines, SMEs are subject to the lower amount. The detail is in the text of Regulation (EU) 2026/1744, which is worth reviewing with your legal team.
Where do I start?
With an inventory of the AI systems in use, including third-party ones, and their classification by risk level. Vivid Vision's pre-assessment does it in weeks, free and with no commitment.
Further reading
Sources
Informational summary, not legal advice.
Reach 2027 with your inventory done
In the pre-assessment we review your AI systems in production, identify gaps against the AI Act and give you a risk map with priorities. Free and with no commitment.
Request your pre-assessmentKeep reading
What is an AI Operating Partner (and how it differs from a consultancy)
What an AI Operating Partner does, how it works and how it differs from an AI consultancy.
Read the pageHow to govern AI in your company: a 4-step model
A four-step model for governing AI: pre-assessment, inventory and risk framework, operation and handover.
Read the page