Brazil Advances AI Consumer Protection Rules With Real Teeth

Technology

Brazil has taken another step toward an AI consumer protection regime, with a congressional committee backing rules that would force companies to come clean about the machines making decisions for people.

The vote, in the science and technology committee of the lower house, is early in a long road. The bill must clear two more committees and both chambers of Congress before it can become law.

Even so, the text is a marker of where Brazil is heading. It pulls openly from the European Union’s AI Act and the country’s own data-protection law to build a rulebook aimed squarely at the everyday user.

What the AI consumer protection bill requires

At its core sits a duty of disclosure. The bill would make companies state clearly when a customer is dealing with an AI system, and explain in plain language what the algorithm does.

The sharper provisions concern automated decisions. When an AI refuses someone credit or returns a medical diagnosis, the consumer could ask for the criteria behind it, challenge the outcome, and demand that a human review the case.

It also creates a right to be forgotten by the machine. Users could ask for their data to be pulled from the pools used to train or run AI systems, with an exception carved out for the credit-scoring world.

One clause looks beyond the usual debates. It would let officially certified AI revalidate prescriptions for long-term medications, a small but concrete sign of how far lawmakers expect the technology to reach into daily life.

Why a foreign reader should care

Brazil is Latin America’s largest economy and one of its biggest consumer markets, so the rules it writes for AI ripple across the region and reach any company selling into it.

The teeth are what set this apart. The bill bans algorithmic discrimination, orders firms to audit their systems for bias, and threatens fines of one to five per cent of revenue, plus a temporary ban on the offending system.

Those duties are continuous, not one-off. Firms would have to run periodic checks for bias and keep open channels for users to complain and seek redress, turning compliance into a standing cost rather than a box ticked once.

It is also a second front. Brazil’s headline AI framework has been stuck in the lower house for months, and this consumer-focused bill is advancing on a separate, quieter track.

The text is a blend. A rapporteur stitched three separate proposals into one and borrowed the architecture from Europe, signalling that Brazil intends to follow the stricter model on offer rather than the lightest.

The contrast with the flagship framework is telling. That wider bill, which would cap penalties in the tens of millions and fight over who pays to train models, has stalled, while these narrower consumer rules keep moving.

For companies deploying AI in Brazil, from banks to hospitals to retailers, the direction is now clear. The country is converging on a European-style model, and the compliance bill that comes with it is taking shape one committee vote at a time.

Frequently Asked Questions

What does the AI consumer protection bill cover?

It requires companies to disclose when consumers are interacting with AI, to explain how the system works, and to allow people to seek human review of automated decisions such as a credit refusal or a medical diagnosis. It also bans algorithmic discrimination and creates a right to have personal data removed from AI training pools.

Is it law yet?

Not yet — a single committee in the Chamber of Deputies approved the text on June 23, and it still needs two more committees and both houses of Congress. The provisions could change as it moves through the process.

What are the penalties?

Firms that break the rules could face a warning with time to fix the problem, a fine of one to five per cent of revenue, or a temporary suspension of the AI system involved. Companies would also have to run periodic bias audits and keep channels for complaints and redress.

This article was drafted with automated assistance and reviewed before publication. How we use AI · Report an error