Europe Has Begun Regulating Artificial Intelligence: How the AI Act Will Change Technology, Business, and People's Lives

On August 2, 2026, the bulk of the provisions of the AI Act—the world’s first comprehensive legislation on artificial intelligence—took effect in the European Union. From now on, European regulators can not only issue recommendations to AI developers but also monitor compliance with the rules, review complaints, request documentation, and impose fines.

Among the most notable changes are the transparency requirements. Chatbots and other interactive systems must inform users that they are communicating with artificial intelligence rather than a live operator. 

Deepfakes must be appropriately labeled, and content created or altered using AI must contain machine-readable tags that help platforms and specialized tools determine its origin.

However, it would be an oversimplification to say that the AI Act became fully effective on August 2. The law is being implemented in phases: certain bans took effect as early as 2025, rules for general-purpose models will take effect in August 2025, and the EU has postponed some of the most complex requirements for high-risk systems until 2027–2028.

UA.News explains what the European AI Act is, which technologies it bans, and how the use of ChatGPT and other generative models will change.

What Is the AI Act and Why Did the European Union Decide to Regulate Artificial Intelligence?

The AI Act, or Artificial Intelligence Act, is a European Union regulation that establishes uniform rules for the development, sale, and use of AI systems on the European market.

The document was finally adopted in 2024 and entered into force on August 1 of that same year. However, European legislation provided for a transition period to allow companies, government agencies, and regulators time to prepare for the new requirements. Therefore, the provisions of the AI Act were introduced gradually, rather than all at once.

The main idea behind the document is not to ban artificial intelligence as a technology. The EU is attempting to categorize AI systems based on the severity of harm they could potentially cause.

A standard spam filter, a video game algorithm, or a recommendation system should not be regulated as strictly as a program that evaluates job candidates, determines a person’s creditworthiness, recognizes faces, or helps law enforcement predict crime.

That is why the AI Act is based on a so-called risk-based approach: the more serious the decision an algorithm can make or recommend, the greater the responsibilities of its developers and users.

What Risk Levels Does the European AI Act Envision?

The European Union broadly divides artificial intelligence systems into four categories: unacceptable, high, limited, and minimal risk.

The first category includes technologies that the EU considers incompatible with fundamental human rights. Their use is prohibited or permitted only in extremely limited, exceptional circumstances.

High-risk systems include algorithms capable of significantly influencing education, employment, access to social services, lending, insurance, immigration procedures, the justice system, healthcare, critical infrastructure, or the work of law enforcement agencies.

Limited risk primarily concerns transparency. People must understand that they are interacting with artificial intelligence or viewing content generated by it.

According to the European Commission’s assessment, the vast majority of everyday applications fall into the category of minimal or zero risk. For these, the AI Act does not impose significant additional restrictions. These include, for example, anti-spam filters, certain recommendation algorithms, and AI features in video games.

Thus, the EU is not attempting to regulate every algorithm. The main focus is on cases where an automated decision could affect the safety, well-being, or fundamental rights of a specific individual.

Which AI Technologies Has the EU Already Banned?

The first bans under the AI Act took effect on February 2, 2025. They target practices that European lawmakers have deemed unacceptably dangerous.

In particular, systems that use covert manipulative or deceptive methods and can significantly alter human behavior are banned. It is also prohibited to exploit people’s vulnerabilities due to their age, disability, or difficult socioeconomic situation.

It is prohibited to create social rating systems in which citizens are evaluated based on their behavior, personal characteristics, or lifestyle, and then have their opportunities restricted as a result.

A separate restriction applies to predicting whether a specific person might commit a crime if the conclusion is based solely on that person’s personality or profiling.

It is also prohibited to indiscriminately collect photographs from the internet or surveillance cameras to create or expand facial recognition databases. In educational institutions and workplaces, with certain exceptions, emotion recognition systems may not be used.

In July 2026, a ban was also added to the legislation on AI systems designed to create sexualized images of people without their consent, including so-called “digital undressing” programs, as well as systems for generating child sexual abuse material. These additional prohibitions are set to take effect on December 2, 2026.

How ChatGPT, chatbots, and generative AI will change starting August 2, 2026

The most noticeable part of the AI Act for ordinary users is Article 50, which regulates the transparency of AI systems.

Starting August 2, chatbots, virtual advisors, voice assistants, and other interactive systems must clearly inform users that they are interacting with artificial intelligence. An exception may apply when this is obvious from the context.

For example, a company must not pass off an automated customer support bot as a real manager. Users must know that responses are generated by an algorithm and take this into account when making decisions.

Providers of systems that generate images, videos, audio, or text must ensure that such content is technically labeled. This labeling must be machine-readable so that digital platforms, search engines, and specialized services can determine the origin of the material.

At the same time, a visible label stating “created using AI” is not required for every use case of a generative model. Special requirements apply to deepfakes and to texts on issues of public importance if they were created by AI and published without human verification or editorial oversight.

If, however, the editorial team used AI to prepare a draft but a journalist fact-checked the material, edited it, and assumed editorial responsibility, the rules may not require the same labeling as in the case of a fully automated publication.

Separately, people must be informed about the use of emotion recognition systems or biometric categorization.

A transition period is provided for certain generative systems that were already on the market before August 2, 2026. They must comply with specific technical requirements regarding machine-readable labeling by December 2, 2026.

What the AI Act Envisages for ChatGPT, Gemini, and Other General-Purpose Models

A separate section of the legislation addresses general-purpose artificial intelligence models, or GPAI. These are large models capable of performing a wide range of tasks: generating text and images, analyzing documents, writing code, translating, summarizing information, or serving as the foundation for other applications.

This category may include the models powering ChatGPT, Gemini, Claude, and other similar services.

The main requirements for GPAI providers took effect on August 2, 2025. Companies must maintain technical documentation, provide the necessary information to developers who integrate the model into their own products, comply with EU copyright law, and publish a detailed description of the materials used to train the model.

Additional requirements have been established for the most powerful models capable of posing systemic risks. Their developers must assess the model’s hazardous capabilities, conduct testing, mitigate risks, report serious incidents, and ensure an adequate level of cybersecurity.

The European AI Office may require providers to submit documentation, conduct model assessments, and compel companies to rectify violations.

For global technology corporations, this means that launching a new model in Europe without explaining how it was tested and what risks it may pose will become increasingly difficult.

What Is Considered High-Risk Artificial Intelligence

The AI Act imposes the strictest practical requirements on high-risk systems.

These may include algorithms used to screen job candidates, evaluate employees, admit students to educational institutions, grade exams, determine creditworthiness, set life or health insurance premiums, reviewing applications for social assistance, or managing immigration procedures.

This category may also include biometric systems, software for critical infrastructure, certain medical devices, and components of machines, vehicles, elevators, and other physical products.

Before such a system is released to the market, the developer must establish a risk management system, verify data quality, maintain technical documentation, ensure that algorithm actions are logged, provide for human oversight, and ensure an adequate level of accuracy, reliability, and cybersecurity.

In many cases, the system must undergo a conformity assessment. If its operation could significantly impact human rights, government agencies and individual companies must also conduct a fundamental rights impact assessment.

However, this particular part of the AI Act has not yet been fully implemented. In July 2026, the AI Omnibus Act took effect, altering the original timeline.

Requirements for systems in high-risk sectors—including education, employment, lending, migration, and law enforcement—will take effect on December 2, 2027. For AI embedded in regulated physical products, the transition period has been extended to August 2, 2028.

The postponement was due, among other things, to delays in developing the technical standards that companies must use to demonstrate their products’ compliance with the law.

Who Will Oversee Compliance with the AI Act

Enforcement responsibilities are shared between the European Commission and the national authorities of member states.

The European AI Office plays a central role. It primarily oversees providers of general-purpose models and the most powerful models posing systemic risk.

The Office may request technical documentation, evaluate models, require corrective measures, and impose financial sanctions. Starting August 2, 2026, oversight and enforcement will also be carried out by the relevant national authorities of EU member states.

In addition, a coordination system has been established, comprising the European Artificial Intelligence Council, a scientific panel of independent experts, and an advisory forum.

Users and organizations may file complaints regarding AI systems. Individuals significantly affected by a decision made on the basis of a high-risk algorithm will, in cases provided for by law, have the right to receive a clear explanation of the role of AI in that decision.

For example, if a person is denied a loan, social benefit, or employment due to the conclusion of a high-risk system, they should be able to find out how the algorithm influenced the outcome.

What Penalties Do Companies Face for Violating the AI Act?

The AI Act provides for fines commensurate with those for violations of European data protection laws.

The most serious violation—the use of prohibited AI practices—can cost a company up to 35 million euros or up to 7% of its total global annual revenue for the previous fiscal year, whichever is greater.

Other violations of the AI Act are also subject to fines in the millions, the amount of which depends on the nature of the violation, the size of the company, the duration of the issue, the level of liability, and cooperation with regulators.

More proportionate approaches are to be applied to small and medium-sized enterprises. The AI Omnibus also extended some of the simplifications to small and mid-cap companies and expanded businesses’ access to regulatory “sandboxes,” where products can be tested under regulatory supervision.

How the AI Act Will Affect Tech Companies Outside Europe

The scope of the AI Act is not limited to companies legally registered in the European Union.

The rules may apply to developers and suppliers from other countries if they offer their AI systems on the EU market, use them within the European Union, or if the results of the system’s operation are used within its territory.

Therefore, American, British, Chinese, Ukrainian, and other tech companies that want to work with European customers must take the AI Act into account.

This could trigger the so-called “Brussels effect.” Large corporations often do not create entirely different versions of products for each market but gradually extend the strictest standards to all countries.

A similar situation has already occurred with the General Data Protection Regulation (GDPR). European privacy rules have influenced the operation of websites, advertising platforms, and digital services far beyond the EU’s borders.

The AI Act could potentially repeat this scenario. Automated labeling of synthetic content, model documentation, risk assessments, and human oversight could gradually become a global standard.

How the New Rules Will Affect Everyday Users

For most people, the AI Act won’t mean the end of ChatGPT, image generators, or recommendation algorithms.

The changes will be less radical but still noticeable. Users will more often see warnings about interacting with AI, labels on deepfakes, and notifications about the use of biometric or emotional systems.

In the long run, people will have more opportunities to challenge decisions in which an algorithm played a significant role. This applies particularly to employment, education, lending, insurance, and government services.

Companies, for their part, will have to better control the data on which their systems are trained or operate. They will not be able to simply claim that “the algorithm decided so” if that decision seriously affected human rights.

The main principle of the AI Act can be summarized as follows: the more power artificial intelligence gains over people, the higher the requirements for its transparency, safety, and accountability must be.

Could the AI Act Slow Down the Development of Artificial Intelligence in Europe?

Critics of the regulation fear that complex documentation, legal risks, and hefty fines could make AI development in Europe more expensive. It will be harder for startups to compete with corporations that have large legal departments and the resources to certify their products.

There is also a risk that some companies will delay the launch of new features in the EU or release them first in markets with less stringent regulations.

It is precisely these concerns that were one of the reasons for the adoption of the AI Omnibus. European lawmakers have extended the deadlines for high-risk systems, simplified some administrative requirements, and expanded opportunities for testing technologies in regulatory sandboxes.

Supporters of the AI Act, on the other hand, believe that clear rules can boost trust in AI and create a predictable market. A company that proves the safety of its product in the EU will be able to use this as a competitive advantage in other countries.

The actual impact will depend not only on the text of the law but also on how national authorities implement it. Overly aggressive oversight could stifle innovation, while merely complying with requirements on paper without actual inspections will not protect people from dangerous algorithms.

The European Union is essentially trying to do with artificial intelligence what it previously did with personal data: establish rules that, due to the size of the European market, will gradually be adopted worldwide.

For the tech world, this means a gradual transition from an era when developers could first release a system and only later consider the consequences, to a model in which risks must be assessed before the product is launched.

It will take a few years to see just how successful this approach will be. But one thing is already clear: Europe was the first to create a comprehensive set of rules for artificial intelligence, and tech companies around the world will have to take it into account.

Read us on Telegram and Sends