AI in Medicaid: New Guides Offer Guardrails for High-Stakes Overhaul

WASHINGTON – May 11, 2026 – A major health coalition today released a detailed playbook for deploying artificial intelligence within state Medicaid programs, aiming to establish critical safeguards ahead of a sweeping legislative overhaul that threatens to strip millions of Americans of their health coverage. The Coalition for Health AI (CHAI) unveiled two Best Practice Guides designed to help states manage a massive new administrative load while preventing the wrongful termination of benefits.

The guidance arrives at a critical juncture. States are grappling with how to implement the complex and controversial community engagement requirements mandated by the H.R.1 legislation, signed into law last year. The new rules are poised to create an unprecedented bureaucratic challenge, and officials are increasingly looking to AI for solutions. CHAI's guides represent a concerted effort by over 40 health organizations to ensure that this turn toward technology prioritizes patient protection over pure automation.

"We convened experts and organizations closest to this work – from community health centers to technology developers – because the people implementing these new Medicaid requirements needed clear and consistent guidance," said Brian Anderson, MD, CEO of CHAI. "The guides reflect nearly a year of rigorous, cross-sector collaboration and provide a clear set of best practices grounded in our responsible AI principles, with patient access, fairness, and human oversight at the center."

The High-Stakes Challenge of H.R.1

The new guidelines are a direct response to the immense pressure H.R.1 places on the Medicaid system. The legislation, passed in mid-2025, mandates that starting in 2027, many low-income adults must document at least 80 hours per month of work, job training, or other community engagement activities to maintain their health coverage. This transforms eligibility from a one-time determination into a continuous, high-frequency tracking process.

Compounding the complexity, the law also requires eligibility redeterminations every six months instead of annually. Policy analysts have warned that this combination will trigger a 'paperwork pandemic' for state agencies and beneficiaries alike. Projections from the Congressional Budget Office estimate that H.R.1 could cause 11.8 million people to lose Medicaid coverage over the next decade, with nearly 5 million of those losses attributed directly to the new work requirements. Many of these are expected to be procedural disenrollments, where eligible individuals lose coverage simply because they fail to navigate the complex new reporting system.

Faced with this operational tsunami, states are exploring AI to automate document verification, send reminders, and adjudicate eligibility. However, without strict guardrails, such systems risk amplifying errors and creating automated pathways to coverage loss, disproportionately affecting the most vulnerable populations.

A Blueprint for Responsible Automation

CHAI's guides directly address these risks by establishing a clear ethical framework. They focus on the two most high-stakes processes: enrollment and eligibility adjudication. The recommendations are unequivocal on several key points.

First and foremost, the guides call for a complete prohibition on fully automated denials and disenrollment. Any adverse action—any decision that would result in a person losing or being denied coverage—must involve a human reviewer. This "human-in-the-loop" requirement is designed as a firewall against algorithmic errors that could have devastating real-world consequences.

Furthermore, the guidance explicitly forbids "default-to-denial" logic. This means an AI system cannot be programmed to automatically deny an application simply because a piece of data is missing or conflicting. Instead, the system must flag the issue for human intervention. The framework also mandates the creation of robust audit trails and confidence thresholds for high-stakes determinations, ensuring that decisions are transparent and reviewable.

"Community Health Centers (CHCs) are on the front lines of care for 52 million Americans, approximately 50% of whom rely on Medicaid," said Kyu Rhee, MD, MPP, President and CEO of the National Association of Community Health Centers (NACHC), a co-chair of the effort. "These guides underscore how AI can help cut administrative burden, navigate new Medicaid eligibility requirements, and strengthen revenue so more time and dollars go back into patient care."

A Collaborative Model for Building Trust

The strength of the new guidelines lies in the broad consensus that underpins them. The development process, which launched in August 2025, was co-chaired by NACHC, Centene, HealthTech 4 Medicaid (HT4M), and Pair Team, and involved a diverse group of over 40 health systems, patient advocates, academic institutions, and technology companies. This cross-sector collaboration was essential for creating a framework that is both technologically sound and ethically grounded.

"Behind every eligibility determination is a human life — a mother trying to keep coverage for her child, an older adult managing chronic illness, a worker navigating instability," said Adimika Meadows Arthur, Founder and CEO of HT4M. "Responsible AI in Medicaid is not simply a technology issue; it is a moral and public trust issue."

The guides also emphasize the need for human-centered design and accessibility. They mandate that any AI-driven system must offer multi-modal, multilingual pathways and adhere to strict accessibility standards, including WCAG 2.2 AA and the Americans with Disabilities Act (ADA). This ensures that individuals with language barriers, low digital literacy, or disabilities are not left behind.

"Frontline care teams and beneficiaries need AI that is accessible to people who may have language or digital literacy barriers, and be reliable and available on the beneficiaries' busy schedules," noted Peter Morrison, Head of Growth at Pair Team. The guides stress the importance of clear pathways for users to escalate issues and connect with a human support agent.

The release of the BPGs is timed just weeks before a June 1, 2026, deadline for the U.S. Department of Health and Human Services (HHS) to release its own guidance on implementing H.R.1. By putting forth a strong, consensus-backed model, CHAI and its partners are aiming to influence federal policy and establish a national standard for how this powerful technology is introduced into a vital public safety net.