AI Act 2026: European rules on artificial intelligence explained for the public, companies and professionals, including the timetable through 2028.
Updated 16 November 2026. The European regulation on artificial intelligence (the AI Act) introduces a risk-based regulatory framework, rather than a general ban on AI. Some obligations already apply, while others will gradually enter into force through 2028. This guide explains what is regulated, who is responsible and what users and companies should know.
AI Act 2026: key obligations now in force
The AI Act in 2026 is not a blanket ban on artificial intelligence. The legal regime depends on the applicable date, the risk category and the company’s role. The timetable must be checked for each individual system.
What is the AI Act and what does the EU aim to achieve?
The AI Act is an EU regulation for artificial intelligence systems, with different requirements depending on their purpose and level of risk. It aims to protect fundamental rights, health and safety, while providing a predictable framework for innovation. Not all automated software necessarily qualifies as an AI system under the regulation, and obligations differ among providers, importers, distributors and organisations using the systems.
Updated official implementation timeline
- 1 August 2024: the regulation entered into force.
- 2 February 2025: general provisions on AI literacy and the initial prohibitions became applicable.
- 2 August 2025: obligations for providers of general-purpose AI models and governance requirements became applicable.
- 2 August 2026: most relevant rules begin to apply, including the transparency obligations under Article 50 and certain enforcement mechanisms.
- 2 December 2026: an important deadline for certain new prohibitions concerning non-consensual sexual content and for the transition of some transparency obligations.
- 2 December 2027: rules for certain high-risk systems listed in Annex III become applicable, under the amended timetable.
- 2 August 2028: deadline for certain high-risk systems embedded in regulated products under Annex I.
The timetable includes legislative changes introduced through the “Digital Omnibus on AI” simplification package; the dates should be checked again before any legal decision is made.
Four risk categories explained
Prohibited practices
Certain uses considered incompatible with fundamental rights are prohibited. The legal list must be read precisely: not every facial-recognition application or classification of people automatically has the same legal status.
High-risk systems
Some applications in areas such as education, employment and critical infrastructure may be classified as high-risk systems if the legal criteria are met. Depending on the role and the relevant application date, requirements may apply to risk management, data, documentation, human oversight and traceability.
Systems with transparency obligations
Certain AI interactions and certain synthetic content must be presented transparently to users. The precise regime differs for chatbots, deepfakes and other artificially generated or modified material.
Other systems
Many productivity tools and everyday applications do not automatically fall into the high-risk category. Other rules remain relevant, however, including data-protection and consumer legislation.
What this means for a company in Romania
- Take stock of the AI systems used or developed and identify the legal role for each one.
- Check the purposes and risk categories, not merely the product name.
- Document providers, data flows, human intervention and oversight mechanisms.
- Prepare staff who use AI and check AI-literacy obligations.
- Review customer communications for applicable transparency obligations.
- Monitor application deadlines and interpretations by competent authorities.
What this means for ordinary users
A chatbot response may be persuasive but inaccurate. For medical, legal or financial information, sources and human verification remain essential. For apparently real images, audio or video, check the original source, the context and any signs of synthetic generation. Transparency matters, but labelling synthetic content does not automatically establish whether the material is true or false.
Are the AI Act, GDPR and copyright the same thing?
No. The AI Act regulates safety, risk and transparency aspects of AI systems. The GDPR sets rules for the processing of personal data, while copyright law pursues other objectives. These frameworks may apply at the same time.
Frequently asked questions
Are ChatGPT and other chatbots banned?
No. The AI Act does not generically ban chatbots. Obligations depend on the system, model, provider, use and applicable provisions.
Does every company using AI need certification?
There is no single certification requirement for every use of AI. The obligations arising from the risk category and the company’s role must be assessed.
Must all AI-generated text be labelled?
There is no simple rule that says “all text, always”. Transparency obligations are defined by Article 50 and the relevant exceptions or conditions.
Why do articles on the AI Act show different deadlines?
Because implementation is phased and some timetables have been amended. Consult the updated official timeline, not an older summary.
Official sources
Note: This is general informational material, not legal advice. Compliance requires an assessment tailored to a system and the organisation’s role.
More analysis on artificial intelligence
Read InfoBuz guides on labelling AI content in 2026, human control of automated systems and a comparison of available AI models.








Discuția cititorilor
Ai o opinie sau o întrebare despre acest articol? Spune-ne ce crezi.
Fii primul care deschide discuția.
Te rugăm să păstrezi un limbaj civilizat și să discuți la subiect. Comentariile spam sunt filtrate automat.