CFM Puts AI Medicine Rules Into Force With No Grace Period

Brazil’s Conselho Federal de Medicina activated Resolution 2.454/2026 on August 26, imposing immediate ethical sanctions on physicians and institutions deploying AI without governance frameworks.

A flat geometric illustration showing structured document panels connected by flowing lines, with small human silhouettes interacting with institutional data grids against an architectural backdrop.

DIGITAL HEALTH · SEPTEMBER 12, 2026 · LATIN AMERICA

Without a grace period, a transition window, or any phased compliance schedule, Brazil’s first national regulatory framework for artificial intelligence (AI) in medicine took effect on August 26. The Conselho Federal de Medicina (CFM), the federal body that governs physician conduct in Brazil, activated Resolution CFM 2.454/2026 — approved on February 11 and published in the Diário Oficial da União on February 27 — after the mandated 180-day vacância period elapsed, according to Exame. Physicians and institutions that had not yet built governance structures were in breach from the moment the calendar turned.

The scale of the compliance challenge is not trivial. According to the TIC Saúde 2025 survey conducted by Cetic.br, Brazil’s internet and technology research center, 18% of the country’s health establishments already use AI. Among units with more than 50 beds, that figure rises to 31%. Generative AI is the dominant modality, present in 76% of AI-using establishments per the same survey. Every one of those deployments now requires a formal governance program, pre-deployment risk classification, continuous audit mechanisms, and documented patient disclosure — or the institution and the attending physician face ethical sanctions, with civil and criminal liability preserved separately.

What the resolution actually requires

The resolution’s scope is deliberately broad. It covers research, development, governance, audit, monitoring, and clinical use of AI systems. Generative AI tools are expressly named — including virtual assistants and software that supports the drafting of clinical documents — which pulls into scope uses that are already routine: consultation transcription, report generation, and clinical-note assistance. The patient disclosure obligation is explicit: when AI plays a relevant supporting role in care, the patient must be informed in clear, accessible language, and the use must be recorded in the medical record. Delegation to AI of the communication of diagnoses, prognoses, or therapeutic decisions without physician mediation is prohibited, according to Exame’s reporting on the resolution text.

Enforcement falls to the Conselhos Regionais de Medicina (CRMs), Brazil’s regional medical councils, not to a dedicated digital-health regulator. That allocation matters operationally: the CRMs have jurisdiction over physician conduct, not over hospitals, clinics, or technology platforms as corporate entities. Flavia Bahia, a professor at FGV Direito Rio, told Exame that the resolution’s principal merit is ending a normative vacuum — ‘before the resolution, the physician who used AI operated in a grey zone: there was no clear conduct parameter, and that is precisely what most exposes a professional to risk.’ That framing is accurate as far as it goes for individual physicians.

The harder jurisdictional question involves institutions. The resolution explicitly extends its requirements to public and private hospitals, clinics, laboratories, telemedicine platforms, and health insurers — entities that the CFM does not traditionally regulate, since its mandate covers physician conduct rather than corporate governance. Legal analyses cited by Exame flag this extension as juridically questionable, noting it could be challenged by any institution sanctioned under that framing. No court has yet tested the point; the matter remains unresolved. Institutions facing enforcement on this basis would likely contest CFM authority before a finding becomes final.

The compliance gap that opened on August 26

Henrique Fabretti, chief executive of Opice Blum Advogados, a law firm that analyzed the resolution, stated that the text ‘represents a structural change for the sector — from now on, it is not enough to contract an AI solution and put it into operation.’ Institutions must implement a formal AI governance program with technical criteria, validation workflows, and permanent control mechanisms, according to Fabretti’s published analysis. Each system must pass a risk assessment before deployment, classified according to its potential impact on patient health and fundamental rights.

The absence of any adaptation period is the operationally consequential detail. Institutions that had been monitoring the regulation since its February publication had 180 days to prepare; those that had not are now exposed without a remediation window. The resolution does not specify a minimum staffing requirement for governance functions, nor does it define the technical standards that a risk classification must meet — gaps that the available record does not resolve and that the CRMs will presumably interpret case by case.

The commercial dimension is already visible. GestãoDS, a healthtech company based in Santa Maria, Rio Grande do Sul, launched its AI agent Helô in the same week the resolution took effect, according to Exame. The product is designed to convert consultation transcripts into structured follow-up plans — precisely the kind of generative AI tool the resolution now governs. The company’s founders, Marcelo Stangherlin and Felipe Ravanello, positioned the product as having been designed with physician supervision built in from the outset, which aligns with the resolution’s supervisory requirements. No independent verification of that compliance claim has been published.

What to Watch

  1. Whether any CRM initiates a formal enforcement action against a hospital or telemedicine platform — which would test the CFM’s jurisdictional reach over non-physician entities and likely trigger a legal challenge
  2. How the CFM or CRMs define the technical standards for risk classification, given that the resolution text does not specify them
  3. Whether the Advocacia-Geral da União (AGU), Brazil’s federal legal office, which separately notified YouTube on September 4 over AI-generated fake-physician videos, coordinates enforcement posture with the CFM or pursues a parallel regulatory track
THE HEALTHSIGNALS BRIEF
A fortnightly PDF compiling our analysis on pharma, medical devices, health policy and epidemiology across Latin America and the United States.
Step 1 of 2 · Free. Work email for instant access. Unsubscribe anytime. By subscribing you agree to receive The HealthSignals Brief by email and accept our Privacy Policy.

The resolution’s deepest structural effect may fall not on physicians but on the vendors supplying them. A hospital’s governance obligation now runs upstream to its AI contracts: procurement, validation, audit, and monitoring requirements must be embedded in vendor agreements or the institution cannot demonstrate compliance. That converts healthtech sales cycles into compliance negotiations — a shift that no published figure yet sizes, but one that every AI vendor with a Brazilian hospital client is now navigating.

CFM Resolution 2.454/2026 — Regulatory Timeline

February 11, 2026
Resolution approved by CFM
February 27, 2026
Published in Diário Oficial da União; 180-day vacância period begins
August 26, 2026
Resolution takes effect; no grace period beyond vacância
Key Distinction: CFM imposed no adaptation period beyond the statutory 180-day vacância. Compliance obligations commence immediately upon effective date.
Source: CFM Resolution 2.454/2026; Diário Oficial da União
THE HEALTHSIGNALS BRIEF

A fortnightly PDF compiling our analysis on pharma, medical devices, health policy and epidemiology across Latin America and the United States.

Built for pharma, market-access, regulatory, policy and investor teams, with the main findings of our gated reports included.

Your subscription could not be validated.
You have successfully subscribed.

Comments

Related