AI, Innovation & Health Business
Can your health insurance plan use AI to deny treatment? What the law says
Jorge Marin
Science Writer

Lawsuits in the US are exposing coverage denials linked to algorithms; in Brazil, doctors and legal experts are debating whether current rules protect patients.
Anyone who has had a test, surgery or medication denied by their health insurance plan knows that feeling of distrust: did anyone really review my case with the attention it deserves?
In the United States, that distrust has turned into fear, and it has found a target: algorithms. Patients and their families are suing health insurers that, according to the allegations, use artificial intelligence to estimate how long a person “should” be treated. As a result, coverage can be cut off even when doctors recommend continuing care.
The concern has already reached doctors’ offices. In a survey released in 2026, the American Medical Association polled 1,000 physicians, and 60% of them fear that insurers’ use of AI will increase prior authorization denials, prior authorization being the approval a plan requires before it will cover tests, surgeries or medications.
In Brazil, the picture is still different. The Associação Paulista de Medicina (APM), the São Paulo State Medical Association, says it has not received reports of automated denials, and two resolutions from the CFM (Brazil’s Federal Council of Medicine) seek to set limits on medical auditing and on the use of AI in health care. The brand-new Patient Rights Statute (Estatuto dos Direitos do Paciente), in effect since April 2026, does not even mention AI.
Given this mismatch, are these rules enough to keep the American model from arriving here without patients even finding out?
Why did US health insurers’ use of AI end up in court?
In the United States, an analytics software program called nH Predict is at the center of lawsuits against health insurers. Developed by the care management company naviHealth, the program estimates how long elderly patients will need post-acute care, such as stays in skilled nursing facilities.
According to the plaintiffs, the tool is nothing more than an algorithm that turns that estimate into a benchmark for ending coverage, overriding doctors’ judgment.
A report released in October 2024 by the US Senate Permanent Subcommittee on Investigations showed that UnitedHealthcare’s initial denial rate for prior authorization requests for post-acute care more than doubled between 2020 and 2022, rising from 10.9% to 22.7%.
That insurer is part of the UnitedHealth group, which acquired naviHealth in 2020. Starting in 2021, the insurer began automating part of its review of these requests to speed it up. In one of the tests, the system sped up the process and increased denials.
In court, one of the landmark cases involves the family of Gene Lokken, 91, who was recovering from a fracture. According to the lawsuit, the insurer cut off coverage sooner than his doctors recommended, based on an algorithmic estimate, and the family paid for his treatment out of pocket until his death, about a year later.
According to the plaintiffs, more than 90% of denials based on the tool were reversed on appeal. UnitedHealth denies having used nH Predict to make coverage decisions. In March 2026, a federal court in Minnesota ordered the company to turn over documents on the tool’s development, use and oversight, but denied access to its source code and training data.
Are there already algorithm-driven denials in Brazil?
In the state of São Paulo, the APM has not yet received formal complaints from member physicians about automated denials. “We have not been receiving questions from colleagues about changes to their treatment decisions or denials of procedures,” says Marun David Cury, the association’s director of professional advocacy. According to him, there have also been no reports of denials issued by non-physicians.
That does not mean, however, that automated denials don’t exist in Brazil. Caio Henrique Sampaio Fernandes, an attorney and partner at the law firm Vilhena Silva Advogados, says signs of them can already be seen in the cases he handles, “but there is rarely direct proof, because the insurer doesn’t say it used a system.”
“The signs are usually these: standardized, identical denials for different clinical cases; very fast responses, with no review of the medical report; generic justifications that don’t engage with the prescription; and the signature of an auditor who never examined the patient,” explains the health law specialist.
The problem is that a sign is not proof. A response that comes too quickly raises suspicion, but it doesn’t guarantee that an algorithm was behind it. “The difficulty lies in proving who, or what, made the decision,” Fernandes sums up.
The CFM, for its part, has rules on auditing and AI, but one of them has been partially suspended by the courts: Resolution No. 2,448/2025, which bars insurers from altering the attending physician’s treatment decisions when they follow clinical guidelines, scientific evidence and the coverage provided. For Cury, the provisions still in force “clearly establish the importance of treating auditing as a specific medical act.”
Another CFM resolution, No. 2,454/2026, treats AI algorithms only as support tools, not as substitutes for the physician, who remains responsible for the final decision. “Neither health institutions nor health insurers can force professionals to follow recommendations generated by algorithms. Physicians have complete freedom to reject them according to their ethical and scientific judgment,” Cury emphasizes.
As he himself notes, the AI resolution sets limits “for physicians and institutions.” How far the CFM’s rules extend to insurers is precisely what is being argued in the courts. For patients, the practical question remains: what rights can they invoke when their plan denies a treatment?
What does the Patient Rights Statute guarantee to those who have been denied treatment?
The Patient Rights Statute (Law No. 15,378/2026) brings together guarantees such as not undergoing any procedure without consent, receiving clear and sufficient information to make decisions, and taking part in choices about one’s own health. The rules also apply to health insurance plans. For Cury, the law “clearly defines patients’ rights, granting in particular autonomy in choosing their treatment.”
For those whose requests have been denied, Fernandes highlights other rights: quality care when it is needed, a second opinion at any stage of treatment, and access to one’s medical records without having to give a reason, with a free copy.
But these rights have a limit. “The Statute was designed for the care relationship, between patients and those who care for them, not for coverage decisions,” Fernandes explains. That is why the law does not address denials, does not require insurers to disclose who made the decision and does not provide for human review.
Algorithms and artificial intelligence don’t even appear in the text.
In addition, the Statute does not establish a specific penalty for violating these rights. A violation is classified as a situation contrary to human rights, but in Fernandes’s assessment, this has a more symbolic than practical effect “for someone who needs treatment tomorrow.” In April, the ANS (Brazil’s National Supplementary Health Agency, which regulates health insurance plans) announced that it will monitor insurers to ensure they respect the rights set out in the law.
Can patients demand to know whether a denial came from an algorithm?
Those who suspect a denial came from a machine already have a tool they can turn to. Under Article 20 of Brazil’s General Data Protection Law (LGPD), from 2018, it is possible to request a review of decisions made solely by automated systems and to demand explanations of the criteria used.
“In theory, patients can demand it. In practice, they face three obstacles,” Fernandes warns. The first is that the rule applies only to decisions made solely by a machine, “and the insurer will always claim there was a human auditor.” It can also invoke trade and industrial secrecy. Finally, the attorney says, “the current wording doesn’t guarantee that the review will be carried out by a person, because that requirement was vetoed in 2019.”
That is why he considers it more effective to combine the LGPD with an ANS rule. Normative Resolution No. 623/2024 requires the plan to communicate a denial in writing, clearly explain the reason and cite the contract clause or regulation that justifies it. “A generic denial is, in and of itself, a flaw that can be challenged,” he adds.
Who is liable for the denial, and what can patients do?
If a denial causes harm to the patient, the insurer pays the bill, even if the decision went through an automated system. Under Brazil’s Consumer Protection Code, the insurer must compensate for the harm even without fault, in line with Precedent 608 of the Superior Tribunal de Justiça (STJ), Brazil’s Superior Court of Justice. “Outsourcing the review to software doesn’t transfer the liability,” Fernandes sums up.
Liability also extends to whoever signs the denial. The physician auditor is personally liable if at fault, “and also before their professional council, if they signed off on a decision without an individual review,” the attorney says. The software vendor, meanwhile, is likely to be pursued by the insurer itself, which can seek reimbursement from it.
According to the attorney, a case can be made that the vendor is also directly liable, as part of the supply chain, “but I’m not yet aware of any settled case law to that effect,” he says.
For patients, the shortest route is still the ANS. Plans are required to cover what is on the agency’s list of mandatory procedures, Cury points out, and, “in cases of doubts or questions about possible denials,” members can file a complaint through the gov.br portal.
The American experience shows that automated denials rarely announce themselves as such. For Fernandes, the case of Gene Lokken makes it clear that “the problem isn’t the technology, but replacing individual clinical judgment with a statistical estimate.” The US court allowed the claims of breach of contract and breach of good faith to proceed because the plan’s documents stated that decisions would be made by clinicians.
The case has not yet been decided on the merits, but it offers a clear lesson for Brazil: “You don’t need to open up the algorithm to hold someone accountable: you just need to prove how it interferes with the decision,” the attorney concludes.



