AI Act

in English: Artificial Intelligence Act (EU AI Act)

basicspreading

A European Union regulation that treats artificial intelligence systems according to their risk and adjusts what is allowed, what is forbidden, and what must be disclosed about them accordingly.

In plain terms

The AI Act is the European Union's common rulebook for artificial intelligence. It mainly regulates not the technology itself, but how it is used: the same model can be harmless in a recipe recommender and pose a serious risk in a credit assessment. An exception applies to large, general-purpose models, which are subject to rules regardless of their use.

It works like traffic rules. You do not need a license for a bicycle, but you do for a car, and for trucks and hazardous goods transport, there are special exams and regular inspections. The greater the potential harm something can cause, the stricter the rules applying to it.

For you as an individual, it mainly grants rights: you must know if you are talking to a machine, if an image or video was artificially created, and if a high-risk AI is making important decisions about you. The obligations fall on developers and on companies and authorities that use AI.

Unlike previous data protection rules (GDPR), the AI Act does not protect your data but mandates the secure and fair operation of systems. Both apply simultaneously.

How it works

Classification: the regulation handles AI uses according to their risk. The literature usually describes this as four levels: unacceptable, high, limited (transparency), and minimal risk. Special rules apply to general-purpose models.
Prohibition: unacceptable-risk uses are banned. Examples include social scoring, whether carried out by the state or private actors, manipulative techniques causing serious harm, and emotion recognition in workplaces and schools, except for medical or security purposes.
Strict conditions: for high-risk systems (such as recruitment screening, credit assessment, educational evaluation, medical devices), risk management, documentation, logging, and human oversight are mandatory.
Transparency: chatbots must indicate that they are machines, and artificially generated images, sounds, and videos (deepfakes) must be labeled. Developers of general-purpose models (such as large language models) must prepare documentation and publish a summary of the training data.
Monitoring and fines: compliance is supervised by national authorities and the EU AI Office. For the most serious violations, fines can reach up to 35 million euros or 7 percent of global annual turnover, whichever is higher.

An everyday example

You apply for a job, and the company uses software to rank resumes. Under the AI Act, this is a high-risk use case. When these rules become applicable, the company must ensure human oversight of the system and inform you that AI is involved in the decision. The same regulation requires that a customer service chatbot state that you are not speaking to a human.

Who is affected

everyoneYou have the right to know if you are talking to a machine or viewing an artificially created image, sound, or video.

entrepreneur, executiveIf your company uses or sells AI, you must assess which risk level it falls into and ensure staff have AI knowledge.

developerIf you build an AI system or agent, the intended use determines whether documentation, logging, human oversight, and compliance checks are needed. → Code

jobseeker, new graduateAI used in recruitment and workplace evaluation is considered high-risk, so people are entitled to more safeguards there. → Work & HR

A common misconception

MythThe AI Act bans or requires licenses for artificial intelligence in Europe.

In realityIt bans only a few specifically dangerous uses; the vast majority of AI systems pose minimal risk and are subject to almost no new obligations.

Limits and risks

The rules come into force in stages, and deadlines have been adjusted along the way, so always check what applies right now.
It does not cover all areas: AI for military, defense, and national security purposes, systems developed exclusively for scientific research, and purely private, non-professional use are excluded.
Risk classification is not always clear-cut, and much depends on detailed rules, standards, and the practice of authorities, which are still evolving.
The existence of the regulation alone is no guarantee: actual protection depends on how effectively national authorities enforce it.

Similar, but not the same

GDPR – GDPR regulates the processing of your personal data, while the AI Act ensures the safety and fair operation of AI systems, even when they do not process personal data.
Digital Services Act (DSA) – The DSA regulates content management and liability of online platforms, while the AI Act covers the AI systems themselves, wherever they are used.
AI ethics guidelines – Ethical guidelines are voluntary recommendations, while the AI Act is mandatory legislation, the violation of which carries fines.

Related concepts

Broader

AI regulation, AI governance, European Union law

Narrower

High-risk AI system, Tiltott AI-gyakorlatok, General-purpose AI model (GPAI), Transparency obligation, AI literacy

On the edge

GDPR, Deepfake, Compliance, Human oversight, Responsible AI

A brief history

The European Commission presented the proposal in April 2021. Regulation (EU) 2024/1689 was adopted in 2024 and entered into force on August 1, 2024. The bans and AI literacy obligations apply from February 2025, while the rules for general-purpose models and the penalty regime apply from August 2025. Most transparency rules apply from August 2026.

Where things stand

By the end of 2026, most bans, rules for general-purpose models, and transparency obligations will be applicable. The establishment of national supervisory systems is progressing at different paces in each member state. The EU Digital Omnibus package schedules requirements for high-risk systems for later, specifically December 2027, and for systems embedded in products for August 2028. Exact dates should be checked in official sources.

Status: 2026 November

In the news

Appears in 112 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Hungarian tech-regulation expert: EU AI Act is over-regulatory but here to stayNov 6, 2026 · Jog
Parts of the EU AI Act to apply in Northern Ireland, but only to AI-integrated goodsNov 4, 2026 · Jog
EU Digital Omnibus changes AI Act deadlines and heads for trilogueNov 2, 2026 · Jog
Five things German firms believe about the AI Act, and what the text saysNov 2, 2026 · Jog
European Health Data Space opens opportunities for AI, while Georgetown finds gaps in AI health equity guidanceNov 2, 2026 · Health
Anthropic watermarks four older Claude models under EU AI Act; xAI to enable Grok API text watermarkingOct 31, 2026 · Agent Age
AI content labelling rules tighten: EU AI Act, UK action, platform policiesOct 26, 2026 · Marketing
EU AI Act: Digital Omnibus pushes high-risk deadlines to 2027 as Article 50 transparency rules applyOct 26, 2026 · Jog

Also known as

AI Act · EU AI Act · MI-rendelet · AI legislation · Artificial Intelligence Act · AI Act · Artificial Intelligence Act · (EU) 2024/1689 rendelet

Sources

European Commission: AI Act – Regulatory framework on AI

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.