Open Medical Policy for AI

We believe AI should be used to make healthcare coverage decisions. It's faster, more consistent, and frees clinicians from administrative burden. This is not a future to resist — it is a future to build.

But AI needs something that doesn't exist today: medical policies that are open, standardized, and machine-readable.

We also believe in open markets — and AI is how we get there. No human can read and compare thousands of medical policies across insurers and lines of business. AI can. But only if the policies are open and structured. Transparent policy isn't just good governance. It's the data layer that lets AI unlock market choice for patients and employers.

These two goals are the same goal. The transparency that lets AI adjudicate fairly is the same transparency that lets markets function. Structured, open medical policy is the infrastructure for both.

Today, coverage rules in health insurance are hidden.

Medical policies are the clinical criteria that determine whether a treatment, procedure, or service is covered — diagnosis requirements, prior treatment failures, lab values, documentation standards. Prior authorization is the process that requires providers to get approval before delivering care, based on those policies. The prior authorization list (PAL) specifies which services require this approval.

These documents are buried in payer websites as PDFs, written in dense clinical and legal prose, inconsistent across payers, and inaccessible to the systems — human or AI — that need to interpret them.

This is the reality of the current system. Medical policies evolved as legal and clinical documents — written to satisfy regulators and justify decisions on appeal, not to enable comparison or computation. No one designed it this way intentionally; it accumulated over decades. But the result is the same: patients can't understand the rules, employers can't compare PA burden across plans, providers spend hours on paperwork instead of patients, and AI systems can't parse policies reliably. The opacity isn't malicious — but it is a barrier to everything we're trying to build.

Transparency changes everything. When policies are public and structured, employers can evaluate access friction before they buy. Patients can understand what's actually covered. Providers can plan treatment knowing what's feasible. AI systems can adjudicate fairly, because they're working against rules everyone can see.

Standardized, open medical policy kills two birds with one stone. It unlocks the market transparency that informed choice requires — and it provides the structured data that AI needs to operate. Today, AI cannot reliably parse coverage criteria locked in PDFs and prose. That's a barrier to the future we want to build.

These are our principles.

On Transparency

I. Medical policies shall be publicly accessible.

Patients, providers, employers, and the systems that serve them have a right to know the rules governing coverage beforecare is delivered — not after a denial. These are not trade secrets. They are the terms of a contract that affects millions of lives. Industry commitments already exist. AHIP members have pledged to make medical policies available online. But this is not enough. Current disclosures typically cover only some policies, for some lines of business, in unstructured formats that vary by payer. True transparency requires more: publication of all medical policies and prior authorization lists, across all lines of business — commercial, Medicare Advantage, Medicaid managed care, self-funded — in standardized, machine-readable formats. The standard must be: if a policy governs a coverage decision, it must be findable, comparable, and computable.

II. Prior authorization requirements shall be published by line of business.

The same insurer may have dozens of PA variations across commercial, Medicare Advantage, Medicaid, and self-funded products. Which services require authorization — and under which criteria — must be explicit for each product a consumer or employer can purchase. This is the foundation of a functioning market.

III. The market shall be able to compare coverage friction.

Premiums and networks are not enough. Employers and individuals must be able to evaluate how hard it is to access covered services — approval rates, documentation requirements, turnaround times, appeal outcomes — before they choose a plan. Opacity is not a competitive advantage. It is a market failure.

IV. AI requires logic to function.

AI systems are entering utilization management now. They will read clinical documentation, match it against coverage criteria, and make or recommend decisions. This is the future we want — but it only works if AI can reliably interpret the rules. Policies locked in PDFs and written in inconsistent prose are not AI-ready. Transparency is not just a market requirement. It is a technical requirement for the AI-enabled healthcare system we are building.

On Standards

V. Medical policies shall be machine-readable.

Prose written for human reviewers is not infrastructure for a modern healthcare system. Clinical criteria must be expressed as computable logic — explicit conditions that humans and machines can interpret consistently. If rules govern a coverage decision, those rules must be extractable and executable.

This requires standards: common schemas for policy structure, consistent formats for clinical criteria, shared data models that work across payers and systems. Healthcare has FHIR for clinical data. Coverage policy needs the same.

VI. Policies shall use standardized clinical terminology.

"Failed two first-line therapies" and "inadequate response to two preferred agents" cannot mean different things to different systems. Policies must use common ontologies — ICD-10, CPT, SNOMED, RxNorm — so criteria are interoperable across payers, portable across systems, and legible to AI.

VII. Decision logic shall be explicit, not implied.

Coverage criteria are not just lists of requirements — they are logical structures. Is it "A and B and C" or "A and (B or C)"? Are there step-therapy sequences, exceptions, overrides? This logic is currently buried in prose, interpreted inconsistently, and invisible to external systems. It must be explicit, structured, and computable.

VIII. Policies shall carry standardized metadata.

Every policy must be tagged with version history, effective dates, applicable lines of business, and the services it governs. When policies change, the change must be traceable. When a decision is made, it must be clear which policy version applied.

On Open Access

IX. Policy content shall be standardized. Access methods may vary.

The substance of coverage logic — the criteria, the terminology, the decision structures — must conform to open standards so that any authorized system can interpret it. How systems query that content may vary by implementation.

X. Licensed clinical content shall not be a barrier to interoperability.

Two vendors — MCG and InterQual — control the majority of clinical criteria content for inpatient utilization management, each holding roughly half the market. Their licensing terms typically prohibit payers from exposing this logic to external systems, locking clinical intelligence inside proprietary platforms and forcing additional purchases to automate what's already been licensed.

This is not intellectual property protection. This is monopolistic control of market infrastructure.

Licensed content must be portable, interoperable, and subject to the same transparency standards as payer-authored policy. If a payer licenses criteria, that payer must be able to use those criteria in any system — and must be able to meet transparency requirements regardless of the content's origin.

XI. Delegated vendors shall meet the same transparency standards.

Payers delegate utilization management segments to specialty vendors (e.g., oncology, pharmacy…) who operate as black boxes within the coverage system. Delegation does not exempt anyone from transparency. All coverage content — whether authored by payers, licensed from MCG or InterQual, or applied by delegated vendors — must meet the same standards of openness, structure, and portability. The accountability follows the decision, not the contract.

On Accountability

XII. Transparency is the foundation for trustworthy AI in utilization management.

AI is already making coverage decisions. This is good. This is what we want. AI can process requests in seconds, apply criteria consistently, reduce administrative burden, and make healthcare more accessible and affordable.

But AI adjudication is only as trustworthy as the policies it interprets — and the decisions it makes.

Transparent policies make AI auditable. Opaque policies make AI a black box at scale.

AI decisions must be subject to the same transparency and accountability as human decisions — and more. Every AI-assisted coverage determination should be traceable: which policy version was applied, which criteria were evaluated, what logic produced the outcome. This must be documented, producible for compliance, and available for regulatory review.

We are not asking for AI to be trusted blindly. We are asking for the infrastructure that makes AI trustworthy.

XIII. Transparency is an industry-wide obligation.

This is not about payers alone. Every participant in the healthcare system must play by new rules:

Payers must publish policies, adopt standards, and enable interoperability — whether they author content themselves, license it, or delegate decisions to vendors.

Content vendors — MCG, InterQual, and others — must make licensed content portable and allow payers to meet transparency requirements without additional licensing barriers.

Delegated vendors must expose the criteria they apply with the same openness required of payers.

AI companies building systems that interpret coverage must design for auditability, traceability, and compliance from the start.

Providers must ensure their clinical documentation and EHR data meet the standards required for AI to function — complete, accurate, and interoperable. Transparency is a two-way street: if coverage decisions must be auditable, so must the clinical inputs.

Regulators must establish and enforce standards that apply across the ecosystem — not just to payers, but to vendors, content licensors, AI systems, and providers alike.

The Stakes

A trillion dollars a year in administrative costs. Tens of millions of denials. A market where no one can see the rules.

Transparency is not an attack on anyone. It is the infrastructure for a market that works