Rebeca Molina's family says a new state run AI bed routing system in Minas Gerais, Brazil, scored her far below her doctors' assessment and kept her waiting five days for an ICU bed.
Rebeca Cardoso Tenente Molina, a 32-year-old psychologist, was admitted to a public hospital in São João Nepomuceno, a small city in the Brazilian state of Minas Gerais, on June 2, 2026, with what doctors initially suspected were gallstones. Her condition worsened over the next four days, with neurological symptoms and internal bleeding. By the evening of June 6 she was dead, within hours of finally reaching an intensive care unit for the most critically ill, 300 kilometers away in the city of Oliveira. Her death certificate lists septic shock, a life-threatening drop in blood pressure and organ failure triggered by severe infection. Doctors are still investigating whether botulism, a rare paralytic illness caused by a toxin, contributed to her death.
What made the case a national story in Brazil was not the wait alone, but who her family says was responsible for it. Her twin sister, Sâmela Cardoso Tenente Furtado, an attorney, told Brazilian broadcaster G1 that the state's new AI-enabled bed-routing system, Core-MG, scored Molina's condition a 6.8 on a 10-point severity scale, far below the ~10 her treating clinicians believed her case warranted. That gap, the family alleges, is what kept her near the bottom of the ICU queue while her organs began to fail.
Core-MG, formally the Central de Operações para Regulação Estadual, was launched on May 19, 2026 by the Minas Gerais State Health Department (SES-MG). It replaced the older SUS Fácil MG system and is now the single, centralized hub through which the state's public hospitals route requests for ICU and other specialty beds. The system uses machine learning to classify each patient's severity and to rank bed offers against a statewide inventory map that is updated three times a day. It pulls patient records from the federal CadSUS and CNES databases, and the state says more than 200 regulatory physicians still review individual cases before a transfer is denied or downgraded.
The family says the review did not save Molina. After days of appeals inside the system, the family obtained a court order compelling her transfer, paid for a private plane, and flew her to Oliveira. She died within hours of arrival.
The state's account, published in the same G1 report and in subsequent state communications, is that Core-MG registered Molina's case the same day it was opened and that the bed she was eventually offered reflected clinical fit and statewide availability rather than a geographic default. SES-MG also says Core-MG has processed 17.6% more daily bed requests and produced 6.1% more daily regulated admissions since launch, numbers the state presents as evidence the new system is expanding, not restricting, access. The state disputes that the algorithm changed the underlying clinical criteria for ICU admission, and disputes that any score from Core-MG is what delayed the transfer.
Gizmodo's English-language writeup of the case, published June 15, 2026, placed the dispute in a wider frame, citing concerns about algorithmic bias in clinical settings and a recent ChatGPT Health emergency under-triage study. That broader literature is real, but it is not what decided whether Rebeca Molina lived or died. What decided it, on the family's account, was a number between 6.8 and 10 generated inside a system that no clinician at the bedside had the authority to override in real time.
"What we saw was that doctors lost the autonomy to decide if a patient is very seriously ill," Sâmela Cardoso Tenente Furtado told G1. The line has become the rallying point for Brazilian commentators asking whether clinical judgment, in any meaningful sense, is still upstream of the algorithm, or merely downstream of it.
Molina's case is a single one, and the causes of her death have not been adjudicated. The 300-kilometer distance to the nearest available ICU is also consistent with the regional ICU scarcity that predates Core-MG and is endemic to large Brazilian states, a structural fact the AI frame risks obscuring. What the case does establish, on the available reporting, is a sequence in which a severity score produced by Core-MG and a treating clinician's bedside judgment diverged by roughly three points on a 10-point scale, in which a court order, not a clinician's override, was the lever that finally moved the patient, and in which no public answer has yet been given about whether the algorithm's call will itself be reviewed, or by whom. That is the accountability gap Molina's case now names, and the one the state's response, so far, has not addressed.