Guides

A guide by Vitor Araujo

AI regulation

The 10 questions people ask most about AI regulation.

Fresco: virtues seated on a bench, among them a reclining Peace, above a row of citizens.
Allegory of Good Government, by Ambrogio Lorenzetti (1338–1339), in Siena’s Palazzo Pubblico. Public domain. Wikimedia Commons

What you will read

01

How is AI regulated around the world?

AI is regulated through laws and policies that vary across countries, alongside international principles on rights, safety and accountability. In the European Union, requirements depend on the risk of the use, with banned practices, stricter obligations for high-risk uses and transparency duties.

Rules are expanding: China has had measures for generative AI in force since 2023, South Korea passed its AI Basic Act in 2024, and the Council of Europe opened a multinational treaty on AI for signature that same year.

  1. Law professor Anu Bradford describes the United States' approach as market-driven, China's as state-driven and the European Union's as rights-driven.Regulation of artificial intelligence · Wikipedia contributors, Wikipedia, 2026
  2. The OECD AI Principles have 47 adherents, including Brazil, and call for respect for human rights, transparency, safety and accountability.OECD AI Principles · OECD, OECD.AI, 2024
  3. The European Commission describes the AI Act as a risk-based law that includes telling people when they are talking to a chatbot and introduces no new rules for minimal-risk uses such as spam filters.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026

What I think: I recommend tracking the rules in each place and, as they develop, evaluating AI before promising results, explaining its limits and making human help available.

02

What is the EU AI Act?

The EU AI Act is the European Union's law that sets rules for artificial intelligence according to the risk of its use. It bans certain practices, imposes strict obligations on high-risk uses and requires transparency in situations such as talking to a chatbot.

Photo of the European Parliament’s semicircular chamber, with hundreds of seats.
The European Parliament’s chamber in Strasbourg. Diliff, CC BY-SA 3.0, colors changed. Wikimedia Commons

The law has been in force since August 2024, but its requirements apply in stages. This matters because a spam filter, a chatbot and a tool for selecting workers are treated differently.

  1. The European Commission describes four risk levels, covering banned practices, obligations for high-risk uses, transparency duties and no new rules for minimal-risk uses such as spam filters.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026
  2. The European Commission's guidelines explain banned practices such as harmful manipulation and social scoring, but are non-binding: authoritative interpretation rests with the EU Court of Justice.Commission publishes the Guidelines on prohibited artificial intelligence (AI) practices, as defined by the AI Act · European Commission, Shaping Europe's digital future, European Commission, 2025
  3. The law classifies AI used to recruit, select and evaluate workers as high-risk, with duties such as human oversight and keeping records of what the system does.Regulation (EU) 2024/1689, the AI Act · European Union, 2024

What I think: I would prepare for the law's next stages by demanding clarity about AI use, human oversight and recordkeeping, especially when selecting and evaluating workers.

03

When does the EU AI Act apply?

The EU AI Act entered into force on 1 August 2024, but its obligations apply in stages, with most applying from 2 August 2026. Rules for high-risk uses start on 2 December 2027, or 2 August 2028 for AI built into regulated products.

There is no single deadline for every company and every use. The first bans and AI literacy duties, which involve knowing how to work with AI, apply from 2 February 2025; general-purpose AI rules apply from 2 August 2025. The timetable depends on the type of AI and how it is used.

  1. The European Commission describes different obligations depending on risk, including telling people when they are talking to a chatbot, and no new rules for minimal-risk AI such as spam filters.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026
  2. According to the European Commission, the AI Omnibus entered into force on 27 July 2026 and set the high-risk deadlines at 2 December 2027 and 2 August 2028, depending on the use.AI Omnibus enters into force · European Commission, Shaping Europe's digital future, European Commission, 2026

What I think: I would use this timetable to evaluate AI before promising safety, rather than treating the later deadlines as a reason to put off testing.

04

What is Brazil's AI bill, PL 2338?

PL 2338/2023 is a bill proposing rules for artificial intelligence in Brazil based on the risk of each use. Under the Senate-approved version, high-risk applications require an impact assessment, and decisions with significant legal effects carry rights to an explanation and human review.

Photo of the two towers and domes of Brazil’s National Congress, designed by Oscar Niemeyer.
Brazil’s National Congress, in Brasília. Agência Senado, CC BY 2.0, colors changed. Wikimedia Commons

The bill could affect AI used in hiring, dismissals, student selection and medical diagnosis. Senate approval did not finish the process: as of 10 November 2026, the text was still awaiting the rapporteur’s report in the Chamber of Deputies’ special committee, before a vote of the full Chamber.

  1. The Senate’s page records that PL 2338/2023 was approved on 10 December 2024 and sent to the Chamber of Deputies on 17 March 2025.Projeto de Lei n° 2338, de 2023 (Marco Legal da Inteligência Artificial) · Senado Federal, Senado Federal, Atividade Legislativa, 2023
  2. As of 10 November 2026, the Chamber listed the bill as awaiting the rapporteur’s report in the Special Committee on Artificial Intelligence, subject to a vote of the full Chamber.PL 2338/2023, ficha de tramitação · Câmara dos Deputados, Câmara dos Deputados, 2026
  3. According to Agência Senado, the approved text bans uses deemed excessively risky, requires impact assessments for high-risk uses and provides for explanations and human review of decisions with significant legal effects.Senado aprova regulamentação da inteligência artificial; texto vai à Câmara · Agência Senado, Agência Senado, 2024

What I think: I recommend that companies prepare for regulation by assessing AI risks and ensuring explanations and human review for decisions that affect people.

05

Does Brazil have an AI law?

Brazil’s AI framework bill is not yet law. Its data protection law already provides rights concerning decisions made solely through automated processing of personal data.

As of November 10, 2026, PL 2338/2023 had passed the Senate but was awaiting a report in the Chamber of Deputies. That debate does not remove existing data protection rights, including the right to request a review of automated decisions that affect a person’s interests.

  1. The Chamber of Deputies records that, as of November 10, 2026, PL 2338/2023 was awaiting its rapporteur’s report in the special committee and remained subject to a full Chamber vote.PL 2338/2023, ficha de tramitação · Câmara dos Deputados, Câmara dos Deputados, 2026
  2. Agência Senado reported in October 2026 that the Senate-approved bill remained under discussion in the Chamber, with 37 proposals attached.Descartada em encontro de EUA e China, regulação da IA é analisada no Brasil · Marcio Maturana (Agência Senado), Agência Senado, 2026
  3. Article 20 of the data protection law grants the right to request a review and clear information about the criteria behind these decisions, subject to trade secrets, but does not require a person to conduct the review.Lei nº 13.709, de 14 de agosto de 2018 (Lei Geral de Proteção de Dados Pessoais), art. 20 · Brasil, Planalto, texto compilado, 2018

What I think: I recommend following the AI framework bill’s progress while already asking for clear explanations of automated decisions and a way to speak with a person.

06

What does Brazil's LGPD say about automated decisions?

Article 20 of Brazil’s data protection law, the LGPD, lets people request a review of decisions made solely through automated processing of their personal data that affect their interests, as well as clear information about the criteria used. The law does not require a person to carry out that review.

The difference between a right to review and a right to human review matters when challenging an automated decision. Transparency also has limits: trade secrets are protected, but if they are used to withhold information, the national authority may conduct an audit to check for discrimination.

  1. Article 20 provides a right to request review of solely automated decisions affecting a person’s interests, including profiling, and to receive clear information about the criteria used, subject to trade secrets.Lei nº 13.709, de 14 de agosto de 2018 (Lei Geral de Proteção de Dados Pessoais), art. 20 · Brasil, Planalto, texto compilado, 2018
  2. In 2019, the President vetoed a human-review requirement, arguing that it would make many companies’ business models unviable and affect credit risk analysis.Mensagem nº 288, de 8 de julho de 2019 (veto parcial ao projeto que originou a Lei nº 13.853/2019) · Presidência da República, Câmara dos Deputados, Legislação Informatizada, 2019
  3. Congress’s record shows that lawmakers upheld the veto of the human-review requirement in October 2019.Veto nº 24/2019, dispositivo 24.19.001 (§ 3º do art. 20 da Lei nº 13.709/2018) · Congresso Nacional, Congresso Nacional, 2019

What I think: I believe companies automating decisions should explain their criteria and provide access to a person to challenge those decisions, without presenting that choice as an LGPD requirement.

07

Does the United States have an AI law?

Yes, the United States has federal and state laws addressing uses of AI, including the Take It Down Act against deepfake abuse and Colorado’s law on automated decisions. The broad federal framework presented by the White House in March 2026 is a proposal to Congress, not a law.

The dispute concerns who sets the rules: the federal government seeks a national standard and wants to challenge certain state laws, while Colorado has passed documentation, explanation and human-review requirements that take effect in January 2027.

  1. A December 2025 executive order directs the creation of a task force to challenge state AI laws in court while a national standard is pursued with Congress.Executive Order 14365: Ensuring a National Policy Framework for Artificial Intelligence · Donald J. Trump, The White House, The White House, 2025
  2. In March 2026, the White House asked Congress for a federal AI framework that would limit certain state laws and build on the already signed Take It Down Act.National Policy Framework for Artificial Intelligence: Legislative Recommendations · The White House, The White House, 2026
  3. Colorado’s SB26-189, signed in May 2026, requires developer documentation, consumer notices and explanations, and the option to request human review starting in January 2027.SB26-189: Automated Decision-Making Technology · Colorado General Assembly, Colorado General Assembly, 2026 Regular Session, 2026

What I think: I recommend tracking which rules apply and when they take effect, prioritizing systems that document their limitations, explain outcomes and allow human review.

08

What is high-risk AI?

Under the EU AI Act, high-risk AI is a category tied to how a system is used, such as screening job applicants, calculating credit scores or forming part of certain products covered by safety rules. This classification brings stricter obligations, but does not mean the use is banned.

The distinction matters because the rules do not treat all AI alike: the AI Act separates banned uses, high-risk uses, transparency duties and minimal-risk applications. In Brazil, the bill approved by the Senate in 2024 also classifies AI by risk level.

  1. The European Commission distinguishes high-risk uses, which face strict obligations, from banned practices and minimal-risk applications such as spam filters.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026
  2. Article 6 includes certain products and safety components covered by EU laws and requiring third-party assessment, as well as the uses listed in Annex III, with some exceptions.Regulation (EU) 2024/1689 (AI Act), Article 6: Classification rules for high-risk AI systems · European Union, AI Act Service Desk, European Commission (consolidated text as at 27 July 2026), 2024
  3. Annex III includes uses in education, employment and essential services, such as screening job applications, credit scoring and pricing life and health insurance.Regulation (EU) 2024/1689 (AI Act), Annex III: High-risk AI systems referred to in Article 6(2) · European Union, AI Act Service Desk, European Commission (consolidated text as at 27 July 2026), 2024
  4. According to Agência Senado, the Senate-approved bill requires algorithmic impact assessments for high-risk uses and provides for explanations and human review of decisions with relevant legal impact.Senado aprova regulamentação da inteligência artificial; texto vai à Câmara · Agência Senado, Agência Senado, 2024

What I think: I would only adopt AI for these uses with evaluation before deployment, clear explanations and a person able to review the decisions.

09

Which uses of AI are banned?

In the European Union, the AI Act bans uses such as manipulation that causes significant harm, social scoring and certain forms of biometric surveillance. Uses classified as high-risk are not automatically banned, but face strict obligations.

Location and timing matter: European bans began applying in February 2025, while the ban on “nudification” apps starts in December 2026. In Brazil, the text approved by the Senate in December 2024 was a bill sent to the Chamber of Deputies.

  1. Article 5 bans predicting crimes solely from personal profiles and recognising emotions at work or school, except for medical or safety reasons.Regulation (EU) 2024/1689 (AI Act), Article 5: Prohibited AI practices · European Union, AI Act Service Desk, European Commission (consolidated text as at 27 July 2026), 2024
  2. The European Commission distinguishes banned practices, high-risk uses, transparency duties and minimal-risk uses.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026
  3. According to Agência Senado, the approved bill bans excessive-risk uses, such as autonomous weapons and social scoring.Senado aprova regulamentação da inteligência artificial; texto vai à Câmara · Agência Senado, Agência Senado, 2024

What I think: I would track the deadlines for bans and measure risks before adopting AI, without reducing people to scores.

10

What does AI regulation change for companies?

AI regulation can require companies to disclose AI use, keep records and provide human oversight, depending on the risk and location. In the European Union, the AI Act sets strict obligations for high-risk uses but introduces no new rules for minimal-risk AI, such as spam filters.

Obligations do not stop with developers: companies that use AI also have responsibilities. Deadlines vary: Colorado’s new requirements start in January 2027, while the EU’s high-risk rules start in December 2027 or August 2028, depending on the case.

  1. According to the European Commission, the AI Act distinguishes uses by risk and includes transparency duties, such as telling people when they are talking to a chatbot.AI Act · European Commission, Shaping Europe's digital future, European Commission (last updated 19 October 2026), 2026
  2. Article 26 provides for companies using high-risk AI to monitor the system, keep automatic logs for at least six months and assign oversight to people with training and authority.Regulation (EU) 2024/1689 (AI Act), Article 26: Obligations of deployers of high-risk AI systems · European Union, AI Act Service Desk, European Commission (consolidated text as at 27 July 2026), 2024
  3. Colorado’s law requires documentation from developers and requires companies using systems for consequential decisions to notify consumers and explain adverse outcomes, while allowing requests for human review.SB26-189: Automated Decision-Making Technology · Colorado General Assembly, Colorado General Assembly, 2026 Regular Session, 2026

What I think: I would prepare for the new obligations by evaluating AI before promising results, documenting its limits and ensuring that a person can intervene.

Who writes this

I’m Vitor Araujo, an engineer. I build AI products for companies, and I speak and write about AI ethics and the philosophy of technology. More about me.

Other guides