Five Countries, Five Models: How BRICS Regulates Artificial Intelligence

The leaders’ statement on global AI governance was adopted on July 6, 2025, at the 17th summit in Rio. Its key ideas include the central role of the UN, consideration of the interests of the Global South, and removing barriers that prevent developing countries from accessing AI knowledge and equipment. The New Delhi Declaration reaffirmed this course.

Russia: Betting on “sovereign” models.

On July 26, 2026, Russia signed Law No. 243-FZ “On Supporting the Development of Artificial Intelligence Technologies in the Russian Federation.” It takes effect on September 1, but its key provisions will come into force on March 1, 2027. The law does not cover all AI, but rather large foundational models—with at least 1 billion parameters—and introduces the concepts of “sovereign” and “national” models. The government will be able to determine the cases in which only these models may be used. The final text is notably softer than the original draft: according to experts, it dropped “trusted” models and a ban on cross-border AI. Labeling AI-generated content remains optional for now; a bill that would make it mandatory for services providing access to large models has been introduced in the State Duma.

China: Targeted rules instead of a single code.

China has no single AI law; its rules are incorporated into existing legislation. As early as 2022–2023, it introduced rules for “deep synthesis” technologies and interim measures for generative AI services. Starting January 1, 2026, the Cybersecurity Law included an article on AI for the first time, combining government support for the industry with safety oversight. Labeling rules have been in effect since September 1, 2025: providers must place visible labels on generated text, images, audio, and video, and embed hidden labels in file metadata.

India: No dedicated law, but tight deadlines.

In November 2025, India drafted AI governance principles that advise against rushing to adopt a separate law. However, amendments to the rules governing IT intermediaries have been in effect since February 20, 2026: synthetic content must be clearly labeled, and illegal material must be removed within hours when ordered by a court or the authorities. The draft amendments also dropped a requirement that the label cover at least 10% of an image. India is also advancing its agenda internationally: the declaration at the February AI Impact Summit in New Delhi was supported by 88 countries and international organizations.

Brazil: The most “European” and the most delayed proposal.
Bill PL 2338/2023, modeled on the European AI Act—with systems classified by risk level and fines of up to 50 million reais—was approved by the Senate on December 10, 2024. Nearly two years later, it is still awaiting the rapporteur’s opinion in a special committee of the Chamber of Deputies. The vote has been postponed several times and is now expected after the October elections.

South Africa: A policy let down by AI.

The Cabinet approved a draft national AI policy on March 25, 2026, and it was opened for public comment on April 10. But in late April, Communications Minister Solly Malatsi withdrew the document after nonexistent sources were found in its bibliography. The minister said the most plausible explanation was that AI-generated references had been included without verification. A revised version is planned for submission to the Cabinet by November 2026.

The bottom line.

China and India are regulating specific risks—above all, deepfakes and content labeling. Russia is betting on supporting its own models. Brazil and South Africa are developing comprehensive frameworks but have yet to complete them. What unites the countries is their foreign-policy position: AI rules should be developed under the auspices of international organizations.

Author: PhD in Economics, Associate Professor, Department of World Economy and International Finance, Financial University under the Government of the Russian Federation, Ruslan Vladislavovich Ozarnov.