Weidner & Friends
← Back to Insights Insights · Recht & KI

EU AI Act: Obligations and Deadlines for Companies

Glass facade of a European institution building at blue hour with EU flags in the wind

The EU AI Act (Regulation (EU) 2024/1689) has applied in large parts since 2 August 2026. For most mid-sized companies this currently means three obligations: building AI literacy among staff, avoiding prohibited practices and making AI visible to customers. The strict rules for high-risk AI apply later, because the Digital Omnibus Regulation (EU) 2026/1744 postponed them: to 2 December 2027 for systems under Annex III, for example in HR, and to 2 August 2028 for AI in products under Annex I. The amendment has been in force since 27 July 2026. This guide shows which AI Act obligations apply to your company and when, with a deadline table, templates and a checklist. As of November 2026.

What the EU AI Act regulates

The AI Act classifies AI systems by risk. At the top are prohibited practices such as social scoring or emotion recognition in the workplace. Below come high-risk systems with extensive obligations, then systems with transparency obligations such as chatbots and AI images, and finally all other applications, for which only the general AI literacy obligation applies. There are also separate rules for general-purpose AI models, which mainly concern the large model providers.

Most applications in mid-sized companies, from text assistants to automated invoice checks, fall into the two lower tiers. That is the good news: the effort is manageable once the basics are in place.

Purpose of the EU AI Act and who it applies to

The Regulation on artificial intelligence is the European Union's first comprehensive AI law. Its purpose under Art. 1: strengthen the internal market, promote human-centric and trustworthy AI and at the same time protect health, safety and fundamental rights without stifling innovation. As a regulation it applies directly in all Member States; a German implementing act only sets competences and supervision.

Under Art. 2 it covers providers placing AI systems on the EU market wherever they are based, deployers established in the EU, and providers or deployers in third countries whose AI output is used in the EU. Exempt are, among others, purely personal non-professional use, military purposes and research and development before placing on the market. For a company this means: as soon as staff use AI applications professionally, the AI Act applies.

Provider or deployer: your role under the EU AI Act

Which obligations apply depends on the role. A deployer is anyone who uses an AI system professionally under their own authority, so almost every company using an off-the-shelf tool. A provider is anyone who develops an AI system or has it developed and places it on the market or puts it into service under their own name.

RoleTypical caseMain obligations
DeployerTeam uses a text assistant, marketing uses AI imagesAI literacy (Art. 4), disclosure of deepfakes (Art. 50(4)), for high-risk Art. 26
ProviderOwn chatbot or agent under own brand for customersAdditionally chatbot notice (Art. 50(1)), marking generated content (Art. 50(2)), for high-risk conformity assessment
From deployer to providerThird-party high-risk system under own name or substantially modified (Art. 25)All provider obligations for that system

Be careful with in-house builds: anyone who develops and offers an agent for customers becomes a provider. How to set up such systems properly is covered in our guide to GDPR-compliant AI agents.

EU AI Act deadlines after the Digital Omnibus

The obligations of the AI Act apply in stages. With the Digital Omnibus, Parliament and Council mainly pushed back the high-risk deadlines in July 2026 and added new prohibitions.

DateWhat appliesLegal basis
2 February 2025Prohibited practices, AI literacyArt. 5, Art. 4, Art. 113
2 August 2025Obligations for general-purpose AI models, rules on penalties (except Art. 101)Chapter V, Chapter XII
2 August 2026General date of application, including transparency obligationsArt. 50, Art. 113
2 December 2026New prohibitions on intimate deepfakes without consent and AI-generated child sexual abuse material; marking of generated content for legacy systemsArt. 5(1)(ba) and (bb), Art. 111(4)
2 December 2027High-risk obligations for Annex III systemsArt. 113(3)(c)
2 August 2028High-risk obligations for AI in Annex I productsArt. 113(3)(c)

For companies this means: preparing for high-risk obligations can wait until the end of 2027, the basic obligations apply today.

AI literacy under Art. 4 of the EU AI Act: what the new version requires

Art. 4 used to require companies to "ensure" the AI literacy of their staff. Since the amendment, providers and deployers take measures to support the development of AI literacy among their staff and other persons working with AI on their behalf. The text states explicitly that no specific level has to be guaranteed. The Commission and Member States are to support SMEs in particular.

So the obligation remains, but it is tied to measures, not to an outcome. In practice, training tailored to the task and prior knowledge is enough, together with a record of who attended and when. A template for the record:

AI literacy record: On 20 November 2026 the inside sales team (8 people) attended the training "AI assistant in the quoting process" (90 minutes). Content: permitted data, checking results, labelling, contact person. Next refresher: May 2027.

How such training is structured is shown on our AI training for teams page.

Prohibited practices and transparency in the EU AI Act

Since February 2025, Art. 5 has prohibited, among other things, manipulative techniques that lead people to harmful decisions, social scoring, untargeted scraping of facial images from the internet and emotion recognition in the workplace and in education, except for medical or safety reasons. From 2 December 2026, systems that generate intimate depictions of identifiable persons without their explicit consent or child sexual abuse material are added. For a company this is mainly a purchasing checkpoint: tools that infer employees' emotions from face, voice or other biometric data are off limits.

The transparency obligations under Art. 50 have applied since 2 August 2026. A chatbot must be designed so that people learn they are talking to an AI unless this is obvious; this is the provider's obligation, which deployers implement in their own channels. Deployers must disclose deepfakes. Providers of generative systems mark generated content in a machine-readable way; for systems on the market before 2 August 2026, by 2 December 2026. A template notice for a chat:

"Hello, you are chatting with our digital assistant. It uses artificial intelligence. If you would rather talk to a person, just type 'agent'."

All cases with further templates are in our guide to the AI Act labelling obligation.

High-risk AI under the EU AI Act: when a system is affected

A system is high-risk mainly when it is used in one of the areas listed in Annex III. For mid-sized companies the most relevant area is HR: AI that filters or evaluates applications, helps decide on promotion or dismissal, allocates tasks based on personal traits or monitors performance. Other areas include critical infrastructure, education, creditworthiness checks and pricing in life and health insurance. Under Art. 6(3), such a system is nevertheless not high-risk if it poses no significant risk, for example because it only performs a narrow procedural task or merely prepares a human decision; the provider must document this assessment, and deployers should ask to see it. Systems that profile natural persons always remain high-risk.

Anyone deploying a high-risk system from December 2027 must, under Art. 26, use it according to the instructions, appoint competent human oversight, monitor operation, keep automatically generated logs for at least six months and inform workers' representatives and affected employees before use in the workplace.

Supervision and fines under the EU AI Act

In Germany, under Section 2 of the KI-MIG, the Federal Network Agency (Bundesnetzagentur) is the competent market surveillance authority unless the act provides otherwise; for regulated financial activities it is BaFin, for AI in products the relevant product authority. The act has applied since 29 July 2026.

Fines under Art. 99 are tiered: up to 35 million euros or 7 percent of worldwide annual turnover for prohibited practices, up to 15 million euros or 3 percent for breaches of most other obligations, and up to 7.5 million euros or 1 percent for incorrect information to authorities. For SMEs the lower of the two amounts applies; since the Omnibus, the same applies to small mid-caps for the two lower tiers. Member States also take the economic viability of these companies into account when imposing penalties.

Checklist: implementing the EU AI Act in your company

Would you like to know which AI Act obligations your tools trigger? Feel free to go through your list with us in a short initial call. How AI moves from pilot into daily operation is covered in moving AI pilots into production. This guide does not replace legal advice in individual cases.

Transparency notice: This article was created with the assistance of artificial intelligence.