Skip to content

The EU AI Act explained for working people

By the KI können editorial team10 min read

Written with AI and checked against the sources by a second AI. As of November 15, 2026. More in the notes on AI.

Since 2024, the EU has regulated artificial intelligence with a law of its own, the AI Act. Most obligations fall on providers and businesses, less on you personally. This guide shows what reaches your working day, as of November 2026. It is an orientation, not legal advice.

Three colleagues go through a checklist at a table with binders and a laptop, a shelf of binders behind them.

Key points

  • The more is at stake for people, the stricter the rules. What counts is what an AI is used for.
  • Some uses are banned, such as reading employees’ emotions at work.
  • AI that screens job applications or helps decide on promotions usually counts as high-risk. The strict rules for it apply from December 2027.
  • Since August 2026, an AI chat has to identify itself as AI, and deepfakes have to be disclosed.
  • Businesses that use AI have to promote their employees’ AI literacy. The law doesn’t require a certificate.

What the AI Act is

The AI Act is Regulation (EU) 2024/1689. It entered into force on August 1, 2024 and applies directly in all EU member states, without each country needing its own law for it. Its rules don’t take effect all at once but in stages, most of them by 2028.

In July 2026, the EU amended the regulation with Regulation (EU) 2026/1744, also known as the “Digital Omnibus”. It postponed deadlines for high-risk AI, reworded the obligation on AI literacy and added two bans.

Who monitors compliance in Germany has been set out since July 29, 2026 in the KI-MIG, the German act on market surveillance and innovation promotion for AI; other member states name their own authorities. The central authority in Germany is the Federal Network Agency (Bundesnetzagentur). It is also the contact point for questions and complaints, unless a specialist authority is responsible, and its AI Service Desk offers help, above all for small businesses.

Who the obligations apply to

The regulation mainly distinguishes two roles. Providers develop an AI system and put it on the market under their own name, such as the companies behind the big AI chats. Deployers use an AI system professionally under their own responsibility, for example a business that uses an AI to select staff. If you use AI purely privately, the regulation gives you no obligations.

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Four levels: what counts is what the AI is used for

The regulation doesn’t regulate the technology as such but how it is used. The same AI can be harmless when it drafts a notice and strictly regulated when it pre-sorts job applications. That is why there are levels by risk:

  • Banned: uses that endanger fundamental rights too much, such as reading emotions at work.
  • High-risk: allowed, but with strict obligations, such as when screening job applications.
  • Transparency: allowed, but must be disclosed, such as chats with AI and deepfakes.
  • Minimal: no special obligations, such as spam filters or spell checkers.

Many everyday tasks with an AI chat fall into the lowest level: drafting e-mails, summarizing texts, building a checklist. The AI Act requires nothing special for these, except that businesses promote their people’s AI literacy (see below). Data protection, copyright and your employer’s rules still apply.

What is banned

Since February 2, 2025, certain AI uses have been banned in the EU. For working life, these matter most:

  • Using AI to recognize the emotions of employees at work or of learners in educational institutions, except for medical or safety reasons.
  • Rating people by their social behavior and disadvantaging them because of it (“social scoring”).
  • Influencing people with covert techniques so that they make decisions that harm them.
  • Deliberately exploiting weaknesses due to age, disability or a difficult social or economic situation.
  • Inferring origin, religion, political opinion or sexual orientation from biometric data such as facial images.
  • Collecting facial images untargeted from the internet or from security cameras to build databases.

From December 2, 2026, two more bans are added: AI that creates realistic intimate or sexual depictions of identifiable people without their consent, and AI that creates depictions of child sexual abuse.

High-risk at work: applications, promotions, performance

AI systems count as high-risk if they help decide important things in people’s lives. The regulation lists them in an annex. For working life, it includes among others:

  • Placing targeted job ads, screening and filtering applications, evaluating candidates.
  • Helping decide on promotion, dismissal or working conditions.
  • Allocating tasks based on behavior or personal traits, and monitoring and evaluating performance or behavior.
  • In education: deciding on admission, evaluating learning outcomes or monitoring exams.

Not every AI in these areas is automatically high-risk. If it only performs a narrowly limited auxiliary task and doesn’t materially influence the outcome, an exception may apply. Whether it does is checked by the provider, not by you at your workplace.

If a business uses a high-risk AI, it has to use it according to the instructions for use, give oversight to people with the necessary competence and monitor its operation. Before using it in the workplace, it has to inform the employees and their representatives, such as the works council (Betriebsrat).

Transparency: when a note about AI is needed

Transparency obligations have applied since August 2, 2026. People should know when they are dealing with an AI or with AI-generated content.

  • Chatting with AI: an AI chat, for example on a website, has to identify itself as AI, at the latest with the first message. According to the Commission’s guidelines, a note only in the terms and conditions or the legal notice is not enough.
  • Deepfakes: realistic-looking AI images, videos or audio of people, objects, places or events that could appear real have to be disclosed.
  • Texts for the public: AI texts that inform the public on matters of public interest need a note, unless a person has reviewed them and holds editorial responsibility.
  • Marking by the providers: AI services that generate text, images or sound have to mark their output in a machine-readable way. For services already on the market before August 2026, this applies from December 2, 2026.

According to the guidelines, advertising copy and product descriptions are mostly not covered by the labeling obligation for texts, unless they make claims about health, safety or the environment, for example. An AI-drafted e-mail to a single customer or an internal text doesn’t count as published. If you are employed, your employer decides how content is labeled.

AI literacy: what businesses have to do

Since February 2, 2025, Article 4 of the regulation has required providers and deployers to take care of their people’s AI literacy. This means the knowledge needed to use AI competently and to assess its opportunities and risks.

The 2026 amendment reworded the obligation. Businesses now have to take measures that promote the AI literacy of their employees and of other people working with AI on their behalf. They don’t have to guarantee a particular level for each individual.

  • According to the Federal Network Agency, the obligation applies regardless of sector and size of business.
  • The law doesn’t prescribe a certificate or a standardized training.
  • Possible measures are training sessions, guidelines, self-study material or short briefings.
  • The Federal Network Agency recommends recording the measures: what, how long, who took part.

Hugo runs a bakery with twelve staff. Two sales assistants draft notices and replies to orders with an AI chat. Hugo gave a half-hour briefing: which AI, what may go in, who reads the texts before they are put up. He noted the date and participants on a sheet of paper. For a small business, that is a sensible start.

What it means for you as an employee

If you are employed, your employer is usually the deployer. It bears the obligations under the regulation, not you personally. Your part is to follow its rules and to check AI results before you rely on them.

  • Ask which AI is approved for work and which data may go in.
  • Don’t use your personal account for company data.
  • Label AI content the way your employer specifies.
  • If AI is introduced at your workplace for applications, rosters or performance reviews, you may ask how it is used. If there is a works council, ask there.

Zoe works in the HR department of a care home. Her employer is considering having an AI pre-sort job applications. She doesn’t decide this alone, but she collects the questions the business should settle first. For this she uses an AI chat without asking for a legal verdict:

Sample instruction
Our care home is considering having job applications pre-sorted by an AI. Assign this to the levels of the EU AI Act and briefly explain why.
Then list at most ten questions we should settle with the provider, the data protection officer and the works council before deciding.
No legal advice and no verdict on whether it is allowed. For each statement, name the article you are referring to so we can look it up.

The answer is a list for further questions, not binding information. Check every article it names yourself in the text of the law. An AI can help classify, but it cannot decide anything bindingly.

What it means for you if you are self-employed

If you use AI for your own business, you are a deployer yourself. For most tasks, such as quotes, website texts or replies to inquiries, this changes little. Watch out for three things:

  1. People who work for you: if temps or freelancers use AI on your behalf, make sure they know how to handle it. A short written rule is a good start.
  2. Chat on your website: if an AI answers there, it has to identify itself as AI right at the start. With ready-made chat services, check whether the notice is visible.
  3. Realistic AI images and videos: if something looks real that isn’t, label it. As a rule, don’t show real people at all without their permission.

Milan is a master electrician with five employees. He has written a page on which AI is used in the business, what may go in and who checks results. The page hangs in the office, and new people read it on their first day. More on quotes, websites and customer data is in the guide Self-employed with AI.

The timeline at a glance

What applies from when (as of November 2026)
DateWhat applies
August 1, 2024The regulation enters into force.
February 2, 2025Bans and AI literacy
August 2, 2025Obligations for providers of general-purpose AI models, such as large language models
August 2, 2026Transparency obligations and most of the remaining rules
December 2, 2026Two new bans; marking for AI services already on the market before August 2026
December 2, 2027High-risk rules, for example for HR and education
August 2, 2028High-risk rules for AI in products with their own EU product rules

How to use AI responsibly in everyday work, with data protection, copyright, labeling and rules for the team, is what you practice in short lessons in the learning app “KI können”. The trial lesson needs no account. What the app itself does with AI is explained under Notes on AI.

Sources

Information as of November 15, 2026. Laws and offers can change; this article is not legal advice.