ProzessX
Academy ↗DEBook a free consultation
Deep Dive · Legal

The EU AI Act for SMEs: what applies when, what to do?

Bans, AI literacy and transparency obligations already apply; the high-risk obligations come later. A factual overview for management teams using AI in their business. As of 10/31/2026.

30 minutes, online, free and non-binding.

In short

The EU AI Act (Regulation (EU) 2024/1689) regulates AI according to its risk level. For most SMEs that merely use AI, this mainly means today: building AI literacy among staff, avoiding prohibited practices, and making it clear when a chatbot or phone assistant is powered by AI. High-risk obligations, for example for personnel selection, apply from December 2, 2027.

Deep Dive

EU AI Act: the overview for SMEs

Since August 2, 2026, the EU AI Act has largely been in effect. Many SME leadership teams are now asking: does this even apply to us if we "just" use a chatbot on our website, an AI phone assistant, or a language model in the office? The short answer: yes, but to a manageable extent. Most obligations in the regulation fall on providers of AI systems. Those who deploy AI in their business have fewer, but clear, duties.

This article lays out what applies as of October 31, 2026, what the so-called Digital Omnibus changed, and which steps make sense. It is based on the text of Regulation (EU) 2024/1689, the amending Regulation (EU) 2026/1744, the information pages of the European Commission, and guidance from the German Federal Network Agency (Bundesnetzagentur).

Important

This article is not legal advice. It gives a general overview as of 10/31/2026. Whether and how an obligation applies to your company depends on the individual AI system and its intended purpose, and in case of doubt belongs in the hands of your legal counsel.

What does the EU AI Act regulate?

The EU AI Act, formally Regulation (EU) 2024/1689, applies directly in all EU member states; no German implementing law is needed for the obligations themselves. The regulation does not govern the technology as such, but its use: the higher the risk an application poses to health, safety and fundamental rights, the stricter the requirements.

It distinguishes mainly two roles. Providers develop an AI system, or have one developed, and place it on the market or put it into service under their own name (Art. 3 No. 3). Deployers use an AI system under their own authority, other than for purely personal, non-professional purposes (Art. 3 No. 4). A trades business that integrates a ready-made chatbot is, as a rule, a deployer.

In Germany, the Bundesnetzagentur is the central market surveillance authority for the AI Act. The basis is the Act on Market Surveillance and Innovation Support for Artificial Intelligence (KI-MIG), which came into force on July 29, 2026. For certain sectors, other authorities remain responsible, such as BaFin for regulated financial companies.

What risk tiers are there?

The regulation works with four tiers. For SMEs, they can be understood as follows:

Four levels, from strict to lenient. Prohibited: unacceptable practices, for example social scoring. High risk: strictly regulated, for example personnel selection. Limited risk: transparency obligations, for example chatbots. Minimal risk: largely unregulated, for example spam filters.

Prohibited

Unacceptable practices, for example social scoring.

High Risk

Strictly regulated, for example personnel selection.

Limited Risk

Transparency obligations, for example chatbots.

Minimal Risk

Largely unregulated, for example spam filters.

Simplified illustration, not legal advice.

Prohibited practices (Art. 5)

Certain applications are banned outright, for example social scoring, manipulative techniques that cause significant harm, or the untargeted scraping of facial images from the internet. Particularly relevant for businesses: emotion recognition in the workplace is banned, except for medical or safety reasons. Software that infers employees' mood from their voice or facial expressions during customer calls has no place in a business.

High risk (Art. 6 with Annexes I and III)

High-risk AI is permitted but strictly regulated. Annex III lists areas of use, including employment and personnel management (such as screening and evaluating applications, decisions on promotion or termination, and performance evaluation) and assessing the creditworthiness of individuals. Annex I covers AI in regulated products such as machinery or medical devices.

Transparency obligations (Art. 50)

Certain systems are subject to disclosure requirements. Anyone talking or writing to an AI should be able to recognize that. AI-generated content should be identifiable as such. Typical examples at SMEs: a chatbot on the website, an AI phone assistant, or generated images or videos in marketing.

Minimal risk

Most everyday applications fall into none of the tiers above, for example a spam filter, AI-assisted receipt capture, or a translation aid in the office. There are no special obligations under the regulation here. The duty to build AI literacy (Art. 4) nonetheless applies to all AI systems.

What has applied since when?

The regulation takes effect in stages. With the Digital Omnibus, the EU changed several deadlines. The Commission proposed it on November 19, 2025, and a political agreement followed in May 2026. The amending Regulation (EU) 2026/1744 was published in the Official Journal on July 24, 2026, and has been in force since July 27, 2026. The timeline under Art. 113 as amended:

  • August 1, 2024: the AI Act enters into force.
  • February 2, 2025: prohibited practices (Art. 5) and the duty to build AI literacy (Art. 4).
  • August 2, 2025: obligations for providers of general-purpose AI models, governance and sanction rules.
  • July 27, 2026: the Digital Omnibus enters into force, including a revised AI literacy obligation.
  • August 2, 2026: general start of application, including the transparency obligations under Art. 50.
  • December 2, 2026: new bans on AI that generates non-consensual intimate depictions of real people or abuse material. Until then, providers of generative systems already on the market before August 2, 2026 must implement the machine-readable labeling required under Art. 50(2).
  • December 2, 2027: obligations for high-risk AI under Annex III, for example in HR (originally August 2, 2026).
  • August 2, 2028: obligations for high-risk AI in products under Annex I (originally August 2, 2027).

Important: the Omnibus postponed the high-risk obligations, it did not abolish them. Bans, AI literacy and transparency obligations remain unchanged. The dates above come from the amended Art. 113 and the European Commission's overview of the AI Act, as of 10/31/2026.

What does AI literacy under Art. 4 mean in practice?

Providers and deployers must take measures to ensure, to their best extent, a sufficient level of AI literacy among their staff and other people who operate AI systems on their behalf. That's how Art. 4 reads as amended by the Digital Omnibus. The measures should take into account prior knowledge, training, context of use, and the people affected. Notably, it is not required that every single person reach a specific level of competence.

In practical terms, for a business that means:

  • Whoever uses which AI tools gets a suitable briefing: what the tool can do, where its limits lie, what data may go into it, and who checks the results.
  • The depth depends on the use. Staff who prepare quotes with AI need something different from IT staff who set up systems.
  • A short internal policy on AI use helps make the rules consistent for everyone.
  • The regulation does not prescribe a particular format. Still, documenting training and the policy in writing is a good idea, so you can show what was done if questions come up.

Support is also available from the state: the Bundesnetzagentur's AI service desk offers information on AI literacy and an online tool for an initial assessment of the risk tier, explicitly aimed at SMEs too.

Does this affect your business too?

In a free initial call, we will look together at what matters most for you.

Deployer or provider: what role does an SME have?

An SME that uses a ready-made chatbot or phone assistant from a manufacturer is, as a rule, a deployer. Most obligations in the regulation, such as technical documentation or conformity assessment for high-risk AI, rest with the provider.

The role can shift, though. Under Art. 3 No. 3, anyone who has an AI system developed and puts it into service under their own name is also a provider. Under Art. 25, a deployer becomes the provider of a high-risk AI system if they put it on the market or into service under their own name or trademark, substantially modify it, or change its purpose so that it becomes high-risk. Example: a general text assistant is modified in-house to evaluate job applications.

Anyone having an AI solution custom-built should therefore clarify the question of roles early and in writing with the vendor. That applies to projects with us as well.

What applies to chatbots and phone assistants (Art. 50)?

People must be able to recognize that they are interacting with AI, unless this is obvious to a reasonably well-informed person anyway. The information must be provided clearly and at the latest at the time of first interaction, and must meet accessibility requirements (Art. 50(1) and (5)). This obligation has applied since August 2, 2026.

Formally, Art. 50(1) is addressed to the provider, who must design the system accordingly. In practice, as a deployer you should still make sure the notice actually appears in your setup, for example:

  • Phone assistant: an announcement at the start of the call, such as "You are speaking with the digital assistant of Sample Company. I am an AI."
  • Website chatbot: a visible notice in the chat window before the first message, not just in the privacy policy.
  • AI-generated images, videos or audio that appear realistic, and AI-generated text intended to inform the public on matters of public interest, must be disclosed under the conditions of Art. 50(4). For text, this doesn't apply if a human reviews it editorially and takes responsibility for it.

A special case is a voice modeled on a real person, such as an employee. Here you should clarify the labeling requirement under Art. 50(4) and the person's consent with your legal counsel.

High-risk AI: why it can affect SMEs too

High risk sounds like it's only for large corporations, but it can quickly affect SMEs. The most common example is applicant management: software that automatically filters, ranks or evaluates job applications falls under Annex III No. 4. The same applies to tools that assign tasks based on personality traits, or evaluate employees' performance and behavior. Assessing the creditworthiness of individuals is listed in Annex III No. 5(b); fraud detection is excluded.

Whether a specific tool is actually high-risk depends on its purpose. Art. 6(3) provides for exceptions, for example for systems that only perform a narrow procedural task. This classification is initially made by the provider.

From December 2, 2027, deployers of high-risk AI have, among others, the following obligations under Art. 26:

  • use the system according to the provider's instructions for use,
  • assign human oversight to people who are qualified, trained and authorized for it,
  • ensure that input data they provide is relevant and representative,
  • monitor operation and inform the provider and authority of risks or serious incidents,
  • keep automatically generated logs for at least six months,
  • inform employee representatives and affected staff before deployment in the workplace,
  • inform affected individuals when the system makes or supports decisions about them.

The postponement creates time. Anyone buying an applicant-screening tool today should still ask now how the provider intends to meet the requirements that apply from December 2027.

What penalties does the regulation provide for?

Art. 99 sets the framework for fines; the exact rules are set by member states. The upper limits:

  • Prohibited practices under Art. 5: up to €35 million or 7% of global annual turnover.
  • Violations of other obligations, including deployer obligations under Art. 26 and transparency obligations under Art. 50: up to €15 million or 3%.
  • False, incomplete or misleading information provided to authorities: up to €7.5 million or 1%.

For large companies, the higher value applies in each case. For SMEs, including startups, the lower value applies (Art. 99(6)). Since the Digital Omnibus, this also applies to small mid-cap companies for the two lower tiers (Art. 99(6a)). These are upper limits, not standard amounts. The assessment takes into account, among other things, the nature, severity and duration of the violation as well as the size of the company.

What should SMEs do now?

These steps give you a structured grip on the topic without turning it into a huge project:

  1. Build an AI inventory: which AI tools does the company use, for what, from which provider, with what data? This also includes features built into existing software.
  2. Clarify your role: are you a deployer for every system, or have you had something built or modified under your own name?
  3. Classify the risk tier: prohibited, high-risk, transparency obligation, or minimal. The Bundesnetzagentur's online tool helps with an initial assessment.
  4. Organize training: build AI literacy appropriate to each role and record who learned what.
  5. Check transparency notices: label the chatbot, phone assistant, and AI content in marketing.
  6. Review contracts with vendors: who is the provider, how are Art. 50 and, in future, high-risk obligations fulfilled, where is data processed, is there a data processing agreement?
  7. Document: briefly record the inventory, classification, training and decisions in writing.
  8. Involve the works council: where one exists, involve it early. AI systems that can capture behavior or performance are subject to co-determination under Section 87(1) No. 6 of the German Works Constitution Act (BetrVG).

How do GDPR and the AI Act relate to each other?

Both apply side by side. The AI Act does not replace the GDPR. As soon as an AI system processes personal data, it still needs a legal basis, information for data subjects, and, where applicable, a data protection impact assessment under Art. 35 GDPR. Art. 26(9) of the AI Act provides that deployers of high-risk AI use the provider's information for this purpose. Art. 22 GDPR remains decisive for automated individual decisions. Anyone who already has their data protection in order can build on a lot of that work.

Bottom line: no reason to panic, but reason to tidy up

For most SMEs, the EU AI Act is today mainly a housekeeping task: know what AI is in use, train employees, and tell customers honestly when they're talking to an AI. Anyone evaluating applicants or employees with AI has more to do before December 2027 and should use the time well. We at ProzessX help with the technical classification and implementation in the business. The legal assessment in individual cases belongs with your legal counsel.

Sources and date

Regulation (EU) 2024/1689 (the AI Act), in particular Art. 3, 4, 5, 6, 25, 26, 50, 99 and 113 and Annex III, as amended by Regulation (EU) 2026/1744 (the Digital Omnibus). European Commission information pages on the AI Act. Bundesnetzagentur, press release of 29.07.2026, and the Act on Market Surveillance and Innovation Support for Artificial Intelligence (KI-MIG). As of 31.10.2026. This article is not legal advice.

Frequently asked questions

Before you reach out

Does the EU AI Act also apply to small businesses?

Yes. The regulation applies regardless of company size to everyone who offers or professionally uses AI systems in the EU. There are relief measures for SMEs, such as lower fine caps under Art. 99(6) and support offerings like the Bundesnetzagentur's AI service desk.

Do we have to label our chatbot or phone assistant?

Since August 2, 2026, people must be able to recognize that they are interacting with AI, at the latest at the time of first interaction (Art. 50). Formally, the obligation is addressed to the provider. As a deployer, you should still make sure the notice is visible or audible in your setup.

Did the Digital Omnibus postpone the obligations?

Partly. Regulation (EU) 2026/1744 has been in force since July 27, 2026. It postpones the high-risk obligations under Annex III to December 2, 2027, and under Annex I to August 2, 2028. Bans, AI literacy and transparency obligations continue to apply.

What do we need to demonstrate for AI literacy under Art. 4?

The regulation requires measures that build AI literacy among staff, but not a specific level for every individual and no specific format. Role-based training, an internal policy, and brief written documentation are good practice.

Is an AI tool for applicant selection allowed?

Yes, but it generally counts as high-risk AI under Annex III No. 4. From December 2, 2027, deployer obligations under Art. 26 then apply, such as human oversight and informing employee representatives and staff in advance. Emotion recognition in the workplace, on the other hand, is banned.

Who enforces the AI Act in Germany?

The central market surveillance authority is the Bundesnetzagentur, based on the KI-MIG, which came into force on July 29, 2026. Other authorities are responsible for specific sectors, such as BaFin for regulated financial companies.

Related topics

Responsible AI and the EU AI ActAI and the works councilAI training for companiesAI phone assistantAI consulting for SMEsKnowledge Hub

Where does your business stand on the EU AI Act?

Talk to us about your AI systems: we help with the inventory, the technical classification and training on AI literacy. We'll get back to you within 24 hours with a proposed time for an initial consultation.

Patrick Sager
Patrick SagerYour contact

By submitting this form you agree to the processing of your data under our privacy policy .