AI Regulation Tracker / Regulation signed
Brazil rules physicians stay responsible for every AI-assisted decision, effective August 26, 2026
Brazil's Federal Council of Medicine has issued binding rules making the treating doctor, not the software, answerable for AI-supported clinical decisions. The rules take effect August 26, 2026, and require doctors to log AI use in the patient record.
Brazil has drawn a clear line on medical artificial intelligence: the machine advises, the doctor decides, and the doctor answers for the result. The Conselho Federal de Medicina (Federal Council of Medicine, or CFM) published Resolução CFM nº 2.454/2026 in the Diário Oficial da União on February 27, 2026. It takes effect 180 days later, on August 26, 2026, and it binds every physician practicing in the country.
The rule is short in concept and firm in effect. Artificial intelligence may be used only as a support tool. Responsibility for clinical, diagnostic, therapeutic and prognostic decisions stays with the treating physician. In the council's own framing, the final decision always rests with the doctor, who retains the authority to accept or reject an AI recommendation based on clinical judgment and professional ethics.
That framing matters because it forecloses a defense that is becoming common as AI moves into clinical settings: that the software recommended it. Under this resolution, an AI suggestion is input, not authority. The physician who acts on it is treated as having made the decision, and the physician who declines to act on it is equally within their rights. The council is regulating the professional, not the product.
What the physician must do
The resolution sets out active duties, not just a statement of principle. A physician must exercise critical judgment over what the software produces rather than deferring to it. A physician must use only compliant systems. A physician must stay current on the limits and biases of the tools in use. And a physician must record in the patient's prontuário (medical record) that AI was used to support a decision. That documentation duty is the operational heart of the rule: it creates a traceable record of where a machine informed care and where a human signed off on it.
Rights the patient keeps
The resolution also affirms a set of patient rights that sit alongside the physician's duties. Patients are entitled to clear information about how their care is delivered, including when AI is involved. They keep the right to a second opinion. Their personal data must be protected. And no experimental intervention may proceed without specific consent. These provisions frame AI as something layered onto the existing doctor-patient relationship, not a substitute for it.
Read together, the duties and the rights point to the same requirement: a documented human decision. The prontuário entry is where the two meet. It gives the patient a basis to ask what informed their care, gives the physician a record of the judgment they applied, and gives any later review a trail to follow. A tool that runs silently in the background, unnamed in the chart, is the situation the resolution is written to prevent.
What it does not do
The resolution does not hand liability to software vendors, and it does not let a physician cite an algorithm as the reason a decision went wrong. It does not approve autonomous diagnosis or treatment by a system acting without a doctor. It is not a device-approval regime; it governs professional conduct, the terms on which a licensed physician may lean on AI. A doctor who follows a flawed AI recommendation without independent scrutiny remains the responsible party.
The cross-border angle
For a US reader, the structural contrast is the point. In the United States, oversight of medical AI is split: the Food and Drug Administration clears AI and machine-learning software as medical devices, while the actual practice of medicine, including how a physician uses such a tool, falls to state medical boards. Brazil consolidates the professional-conduct side into a single national instrument. One council, one rule, every physician in the country. The resolution does not bind US doctors, but it is a working model for how a national professional body can govern AI use directly, and it is the kind of standard that US state boards and specialty societies may look to as they weigh their own guidance.
The near-term signal for anyone running an AI-assisted clinical operation touching Brazil is straightforward. From August 26, 2026, the compliance question is not only whether a tool works, but whether the physician using it applied independent judgment and wrote it down.
Frequently Asked Questions
What did Brazil's CFM actually change with Resolução 2.454/2026?
It made binding that AI in medicine is a support tool only and that the treating physician stays ultimately responsible for clinical, diagnostic, therapeutic and prognostic decisions. Physicians must also record AI use in the patient's medical record. It was published February 27, 2026 and takes effect August 26, 2026.
Who has to comply, and by when?
Every physician practicing in Brazil, across all specialties. The rule takes effect August 26, 2026, which is 180 days after its February 27, 2026 publication in the Diário Oficial da União.
Does using AI reduce a doctor's liability if something goes wrong?
No. The resolution keeps ultimate responsibility with the physician. AI is advisory. A doctor who accepts an AI recommendation without applying independent critical judgment remains the responsible party.
What must a physician document?
The physician must record in the patient's prontuário (medical record) that AI was used to support the decision. The resolution also requires using only compliant systems and staying current on the tools' limits and biases.
Sponsored Training
Browse the full AI Regulation News tracker
Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.