EU AI Act for Brazilian Companies: What to Check if You Sell in Europe

EU AI Act for Brazilian Companies: What to Check if You Sell in Europe

The EU AI Act may reach your company if you serve customers in Europe. Learn how to check its scope and which usage records help show compliance.

•

Want to apply this in your company?

Book a call with Juan →

The EU AI Act is the European regulation that classifies artificial intelligence systems by their use and their risk. It can reach a Brazilian company that uses AI to serve customers in the European Union, even without a headquarters or office there. The useful question is not how big your company is or whether the technology is called superintelligence. The useful question is whether any AI system produces outputs that people in Europe rely on, and what kind of decision that system feeds.

Key takeaways

  • The EU AI Act does not use superintelligence as a category. It works with prohibited practices, high-risk systems, transparency obligations and general-purpose AI models.
  • A company outside Europe can fall within scope when the system is placed on the European market or when its output is used inside the European Union.
  • Systems that talk directly with people must, as a general rule, make clear that AI is involved, including on WhatsApp, a website or email support.
  • Using AI in decisions about people, such as credit, hiring or access to essential services, requires a stricter assessment and human oversight.
  • The records that support European compliance look a lot like what LGPD expects from anyone processing personal data in Brazil: an inventory, a stated purpose, logs, tests and named owners.

The EU AI Act looks at use, not at superintelligence

The debate about superintelligence takes up a lot of media space, but the text of the AI Act works with legal categories. It separates prohibited practices, high-risk systems, systems with transparency obligations and general-purpose AI models, and each one has its own rules. A company that uses an assistant to answer questions about orders does not need to decide whether the model is superintelligent. It needs to answer what the system does, who it talks to and what effect it has on the person on the other side.

If your team still mixes the technical concept with sales language, read what ChatGPT is and what it is not before discussing the regulation with your leadership.

When the regulation reaches a Brazilian company

The AI Act has extraterritorial reach. In general terms, it can apply to providers that place systems on the European Union market, and to deployers located outside the Union when the output produced by the system is used in the Union. For a Brazilian company selling services to European customers, the case that deserves attention is the second one: the system runs in Brazil, but the answer, recommendation or decision reaches someone in Europe.

This reading depends on the official text and on the facts of each case. Treat it as a starting point for a check with a lawyer, not as an automatic conclusion.

Checklist: how to tell whether you are within scope

Before drafting any document, do a direct review of your operations. Follow this sequence:

  • List every AI system in use, including the ones staff adopted on their own, such as text generators, support assistants and screening tools.
  • For each one, note who is affected, whether those people are in Europe and whether the output is used there.
  • Identify whether the system talks directly with people, generates published content or supports a decision about them.
  • Check whether any use falls into a sensitive area, such as credit, hiring, education, health or essential services.
  • Mark which uses are automated decisions and which are suggestions that a person reviews before acting.
  • Take the list to a lawyer with experience in technology and data protection to confirm the classification of each item.

Which usage records help demonstrate compliance

Compliance is shown with records, not with promises. These are the ones worth keeping from the start:

  • An inventory of systems, with name, vendor, version, purpose and internal owner.
  • A description of permitted and prohibited uses, with clear limits for the AI and for the team.
  • The data sources used, the legal basis for processing them and an indication of which data belongs to European customers.
  • Records of relevant interactions, with date, input, output, action taken and human intervention when it happens.
  • Assessments and tests run before go-live and during periodic reviews, with success criteria and recorded results.
  • A log of incidents, complaints and corrections applied, with dates and owners.
  • Notices to people informing them that they are talking with AI and how to request a human review.

These records serve more than one regulation. LGPD also covers data processing and automated decisions, so a well-built inventory addresses both fronts. If your company already keeps decision logs, adapting is usually a matter of completing fields and expanding human review, not starting from scratch. For the internal usage policy, the same logic applies to company policy for employee ChatGPT use.

Personal data, LGPD and transfers outside Brazil

When customer service involves people in Europe, another regulation usually appears alongside the AI Act: the GDPR, the European data protection regulation. In Brazil, LGPD governs the processing of personal data, including when that information is sent to AI providers that process it outside the country. For that reason, your data flow map needs to show where each piece of information enters, where it is stored, who processes it and how long it is kept. The details of this point are covered in international data transfer under LGPD when using AI providers.

It is also worth following the debate on Brazil's AI legal framework, because it may create its own obligations for domestic companies. Brazil AI regulation: what companies should watch and do now summarizes what your company should track.

Illustrative scenario: a language school with students in Portugal

Illustrative scenario: a Brazilian language school has students in Portugal and uses an agent on WhatsApp to answer questions about class schedules and guide people through enrollment. The agent does not approve scholarships or reject students. It collects details, answers what is in the school's official timetable and passes everything else to the administration office. In this setup, the team records what the agent can answer, when it hands off to a person, who reviews the AI notices and which data is stored and where. With that in place, the conversation with the lawyer stops being abstract: the company knows what the system does, for whom, within which limits and with which records.

In the method we follow, the path is to map the current process and its baseline, limit data and actions, pilot a real flow with records and an agreed metric, and only then move the operation over and decide whether to expand, adjust or stop. Applied to compliance, the pilot starts with a single system that serves European customers and produces the inventory, the notices and the logs before any expansion. The legal decision on classification remains with your company and your lawyer.

Frequently asked questions

Do I need to follow the AI Act if my company has no office in Europe? It may be required, depending on how your AI system is used and on the output that reaches people in the European Union. The test is not only where the company is registered, but where the system is placed on the market or where its output is used. That is why the check starts with an inventory of your systems and of the customers you serve, followed by a review with a lawyer.

What is superintelligence for the purposes of the AI Act? The regulation does not define superintelligence as a legal category. It classifies systems by use, by risk and by the type of obligation they create, such as transparency, high risk or general-purpose models. For your company, the practical question is what the system does and who it interacts with, not the label a vendor uses in its marketing.

Does a WhatsApp chatbot fall under any obligation? As a general rule, systems that interact directly with people must make clear that AI is involved, unless this is obvious from the context. The service must also respect personal data rules and the policies of the channel itself. It is worth recording the notice, the route to a human agent and the conversations that require review.

Where should I start my compliance records today? Start with an inventory of AI systems, covering purpose, owner, data processed and customers affected. Then define what each system can and cannot do, record the relevant interactions and agree on the tests that prove the expected behavior. This set is simple to maintain and serves both the AI Act and LGPD.

Sources

EU AI ActLGPDAI complianceData privacyCross-border AI