Skip to content
AI governance

AI Act for companies: what it requires and deadlines (Dec 2027)

· 7 min read · Vivid Vision

The AI Act (Regulation (EU) 2024/1689) is the world's first comprehensive law regulating artificial intelligence. It classifies AI systems by risk level and requires the companies that develop or use them to inventory them, manage their risks, document them and guarantee human oversight. It has been in force since 1 August 2024, with phased application. Regulation (EU) 2026/1744 postponed the obligations for high-risk systems to December 2027 and August 2028, but the prohibitions, AI literacy, the rules for general-purpose models and transparency already apply. If your organisation already uses AI in production, preparation starts now.

This guide summarises, without legal jargon, what the regulation requires of a Spanish company, which deadlines are critical and where it makes sense to start.

What the AI Act is and who it applies to

The AI Act is a directly applicable European regulation: it does not need to be transposed into national law to be binding. It applies to any organisation that places on the market, puts into service or uses AI systems in the EU, regardless of where it is established. In other words, it covers both those who develop the AI (the provider) and those who deploy it in their operations (the deployer).

In practice, this includes most companies: a bank that uses credit scoring models, an insurer that automates claims, an HR department that screens CVs with AI or an industrial company with predictive maintenance all fall within its scope.

The four risk levels

The regulation is proportionate to risk. The greater the potential harm of a system, the more obligations it imposes:

  • Unacceptable risk (prohibited): banned practices such as government social scoring, subliminal manipulation or emotion recognition in the workplace and in education.
  • High risk: systems that affect safety or fundamental rights —credit, employment, education, life and health insurance, critical infrastructure, biometrics—. This is the category that accumulates the most obligations.
  • Limited risk: systems with transparency obligations, such as chatbots or AI-generated content, which must identify themselves as such.
  • Minimal risk: most applications (spam filters, recommenders) have no specific obligations.

Key deadlines you cannot ignore

Application is progressive. These are the dates that set the agenda for any management team:

  • 2 February 2025 — Ban on unacceptable-risk practices and the obligation to ensure AI literacy among staff (Regulation (EU) 2026/1744 reworded it as a duty to take measures that promote it).
  • 2 August 2025 — Obligations for general-purpose AI (GPAI) models, such as large language models.
  • 2 August 2026 — General date of application. Since then, among others, the transparency obligations of Article 50 apply, such as identifying chatbots and AI-generated content.
  • 2 December 2026 — The new prohibited practices added by Regulation (EU) 2026/1744 apply, and the deadline to mark synthetic content from systems placed on the market before August 2026 expires.
  • 2 December 2027 — Obligations for Annex III high-risk systems, such as those used in employment, education, credit or biometrics. This is the milestone around which compliance work is concentrated.
  • 2 August 2028 — Obligations for high-risk systems embedded in products that are already regulated (Annex I).

These dates come from Regulation (EU) 2026/1744, known as the Digital Omnibus on AI, published in the Official Journal of the EU on 24 July 2026 and in force since 27 July. The high-risk deadline has moved; the obligations have not. What already applies remains in force, and anyone who waits until 2027 to start will be late.

What obligations it imposes on high-risk systems

If you operate a high-risk system, the regulation requires, among other things:

  1. A continuous risk management system throughout the entire life cycle.
  2. Data governance: representative training and validation datasets free from undue bias.
  3. Technical documentation and records (logs) that enable traceability.
  4. Sufficient transparency for the deployer to understand and use the system correctly.
  5. Effective human oversight of the system's decisions.
  6. Accuracy, robustness and cybersecurity appropriate to the intended use.

For deployers (most companies that use third-party AI) there are specific obligations: using the system in accordance with the instructions, maintaining human oversight, monitoring its operation and retaining the logs.

Penalties: up to €35M or 7% of turnover

The AI Act takes non-compliance seriously. Article 99 sets fines of up to €35 million or 7% of worldwide annual turnover (whichever is higher) for using prohibited practices; up to €15M or 3% for breaching the remaining obligations; and up to €7.5M or 1% for supplying incorrect information. For a medium-sized company, the percentage of turnover can be more painful than the absolute figure.

The Spanish context: AESIA and national legislation

Spain was the first EU country to create a dedicated authority: the Spanish Agency for the Supervision of Artificial Intelligence (AESIA), based in A Coruña, which will oversee the application of the regulation within its territory. In addition, the Government is processing a draft law for the good use and governance of AI that complements the European framework. On top of this, for the financial sector, there is the DORA regulation on digital operational resilience, fully applicable since January 2025.

How to prepare your company (without paralysing the business)

The usual trap is to treat the AI Act as an isolated legal project. It is not: it is an operational change. Sustainable compliance comes from an AI governance framework that lives inside the organisation, not from a report that gets filed away. A realistic starting point:

  1. Inventory. Map all the AI systems that are already making decisions in production —including third-party ones and shadow AI—.
  2. Risk classification. Determine which ones fall into high risk under the regulation.
  3. Gap analysis. Compare each system against the applicable obligations.
  4. Operating model. Assign responsibilities (roles, committee, an AI lead or CAIO), define policies and controls, and embed it in day-to-day work.

At Vivid Vision we don't hand over a PowerPoint and leave: we operate the governance model from the inside, with your team, in a vendor-agnostic way and leaving the knowledge installed. We measure it in NPS, P&L and adoption curve, not in deliverables. — Rafael Tamames, founder of Vivid Vision (LinkedIn Top Voice in AI, founder of Findasense).

If you want to know where to start with your systems, our starting point is an AI governance pre-assessment: we map your systems in production, their risks and your compliance priorities against the AI Act and DORA.

Frequently asked questions

When does the AI Act come into force?

The AI Act came into force on 1 August 2024, but its application is phased. Prohibited practices and AI literacy apply from February 2025, obligations for general-purpose models from August 2025 and transparency obligations from August 2026. Regulation (EU) 2026/1744 postponed those for high-risk systems to 2 December 2027 (Annex III) and 2 August 2028 (Annex I). The deadline has moved, not the obligations, so preparation needs to start with time to spare.

Which companies does the European AI Regulation apply to?

It applies to every organisation that develops, markets or uses AI systems in the European Union, whether or not it is established there. It covers both the provider that creates the system and the deployer that uses it in its operations, which in practice includes most companies that use AI in banking, insurance, healthcare, HR or industry.

What are the penalties for breaching the AI Act?

Fines reach up to €35 million or 7% of worldwide annual turnover —whichever is higher— for using prohibited AI practices. For other infringements, up to €15M or 3%, and up to €7.5M or 1% for supplying incorrect information to the authorities, under Article 99 of the regulation.

What is a high-risk AI system?

It is a system whose use can affect people's safety or fundamental rights: credit scoring, personnel selection, life and health insurance, education, biometrics or the management of critical infrastructure. These systems concentrate most of the AI Act's obligations: risk management, documentation, human oversight and traceability.

Do I need an AI lead in my company?

If you already have AI systems making decisions in production, you need to assign executive responsibility for their governance and risk, whether a dedicated Chief AI Officer (CAIO) or an equivalent role. The AI Act requires human oversight and accountability, which in practice calls for a clear figure who answers for compliance and the operation of AI.


Related reading: ISO 42001, the AI management standard — the framework that makes it easier to demonstrate compliance with the AI Act —, how to set up an AI governance framework and what AESIA is, the authority that enforces the AI Act in Spain.

Sources: the official text of Regulation (EU) 2024/1689 (AI Act) on EUR-Lex, Regulation (EU) 2026/1744 on EUR-Lex and the Spanish Agency for the Supervision of Artificial Intelligence (AESIA). This guide is for information purposes and does not constitute legal advice.

Is your AI already in production without a governance framework?

Start with a pre-assessment: we map your AI systems, their risks and your priorities against the AI Act and DORA. We operate the governance with you, from the inside.

Request your pre-assessment
Request your pre-assessment

Choose which cookies you accept. You can change your choice or withdraw your consent at any time from Cookie settings, in the footer of every page.

Needed for the site to work and to remember your cookie choice.

They tell us how many people visit the site, which content is most interesting and how you found us if you write to us. Used by Google Analytics, Microsoft Clarity and Ploy analytics.

Cookie policy