When a payer drafts a coverage policy for a new specialty drug, the result is supposed to reflect the FDA-approved indication and the available evidence. In practice, payer policies routinely impose criteria that are narrower than the label: age cutoffs that the label does not contain, step-therapy requirements that the label does not mandate, and exclusion criteria that mirror the clinical-trial protocol rather than the prescribing information. What is changing in 2025 and 2026 is that AI and automation tools are now generating these policies at scale, embedding restrictive criteria into electronic prior-authorization systems faster than manufacturers can monitor and respond.
This article is for market-access directors, payer-strategy leads, and medical-affairs professionals responsible for ensuring that coverage criteria reflect the approved label. It maps the mechanisms by which AI-generated policies become more restrictive than the label, the operational risks for launch teams, and the monitoring and response strategies that work.
The problem: payer criteria versus FDA labels
The label-access gap is not new
Pharmaceutical Executive documented in 2025 that payers "routinely write coverage criteria that effectively limit access to only those patients who would have been eligible for the pivotal clinical trial, rather than embracing the broader population encompassed by the FDA-approved label." This creates what the article calls a "profound disconnect between regulatory approval and real-world access."
A 2024 study published in PMC analyzed coverage policies for 187 specialty drugs across 17 large US commercial health plans from 2017 through 2021. The proportion of policies that were more restrictive than the FDA label increased from 39.5% to 51.7% over the study period. The proportion of policies consistent with the FDA label declined from 57.1% to 45.1%. For 13 of the 17 plans, the share of policies with restrictions increased over time. The study documented that patient subgroup restrictions, step therapy protocols, and prescriber requirements were the three most common restriction types imposed beyond the label.
A separate analysis by the Tufts Center for the Evaluation of Value and Risk in Health found that 28% of coverage policies for accelerated-approval drugs included restrictions beyond the FDA label, with subgroup restrictions being the most common type. Oncology drugs were covered with label-beyond restrictions less often (21%) than non-oncology drugs (82%), and orphan drugs were more often covered with restrictions (32%) than non-orphan drugs (20%).
The American Journal of Managed Care reported that 90% of patients with specialty drug prescriptions face some form of utilization management—prior authorization, step therapy, or quantity limits—and that the share of decisions with no utilization management dropped from 69% in 2017 to 37% in 2019.
What is different with AI-generated policies
Three shifts are occurring simultaneously:
Speed and volume. AI tools can draft coverage policies, prior-authorization criteria, and step-therapy protocols in minutes rather than weeks. MACPAC's February 2025 analysis of automation in the Medicaid prior-authorization process identified payer-side AI use cases including triage of incoming PA requests, real-time PA decisions, identification of services for reduced PA requirements, and detection of incorrect or fraudulent PA claims.
Training-data bias. AI models that draft coverage criteria often train on structured data from clinical-trial protocols, published literature, and existing payer policies. Clinical-trial inclusion and exclusion criteria are inherently narrower than the FDA label. FDA's December 2025 final guidance, Enhancing Participation in Clinical Trials—Eligibility Criteria, Enrollment Practices, and Trial Designs, explicitly acknowledges that "overly restrictive inclusion criteria are increasingly linked to limited generalizability, reduced label expansion opportunities, weakened payer and clinician confidence, and lower external validity of trial results."
Opacity. When a human pharmacist or medical director at a PBM writes a coverage policy, a manufacturer can identify the author, understand the rationale, and engage in a dialogue. When an AI system generates the policy, the decision logic may be embedded in a model that the payer itself cannot fully explain. MACPAC flagged "limited transparency" and "potential for bias" as key challenges of automation in the PA process.
Where the tightening happens
Inclusion criteria that mirror the trial, not the label
The most common mechanism by which payer policies become more restrictive than the label is the importation of clinical-trial inclusion criteria into coverage criteria. Examples include:
- Age restrictions. The label may state "adults and pediatric patients aged 12 years and older," but the payer policy may restrict coverage to patients aged 18 to 65, reflecting the age range of the pivotal trial rather than the approved indication.
- Comorbidity exclusions. The label may not exclude patients with hepatic impairment, but the payer policy may require normal liver function, because the pivotal trial excluded patients with AST or ALT above 2x ULN.
- Prior-therapy requirements. The label may state "treatment of moderate-to-severe disease," but the payer policy may require failure of two prior therapies, because the pivotal trial enrolled patients who had failed prior treatments.
FDA's guidance on broadening eligibility criteria notes that these restrictions "may result in the trial not fully representing real-life patients who will receive the experimental drug, if it is approved" and that the same problem extends to payer policies that adopt trial criteria as coverage requirements.
Step-therapy requirements that exceed label recommendations
Payers frequently impose step-therapy requirements that require patients to fail one or more therapies before the new drug is covered, even when the label does not specify a particular treatment sequence. For specialty drugs, step-therapy requirements are often driven by cost rather than clinical evidence, but AI tools that automate policy generation may encode these requirements without distinguishing between cost-driven and evidence-driven criteria.
Quantity limits below labeled dosing
Payer policies may impose quantity limits or dosing maximums that are below the labeled dose. For biologics with weight-based dosing, this can create situations where patients who require higher doses (based on body weight) are denied coverage for the full prescribed amount.
Biomarker requirements not in the label
For oncology drugs, payers may require specific biomarker testing that goes beyond the labeled companion diagnostic requirement. The label may require a specific genetic test, but the payer policy may add additional biomarkers based on clinical-trial subgroup analyses, effectively restricting coverage to a narrower patient population.
The CMS regulatory backdrop
The 2026 Interoperability and PA Rule
CMS's proposed 2026 Interoperability Standards and Prior Authorization for Drugs rule extends the requirements of the 2024 final rule (CMS-0057-F) to cover prior authorizations for drugs. Beginning in 2026, impacted payers must respond to PA requests within 72 hours for urgent requests and 7 calendar days for standard requests. Payers must also post PA metrics for calendar year 2025 on their websites by March 31, 2026.
These requirements are accelerating payer adoption of AI tools for PA processing. The MACPAC analysis notes that payers are deploying "predictive AI" and "large language model input for unstructured data" to triage incoming PA requests and generate real-time decisions. CMS's proposed rule also requires impacted payers to make detailed PA decision information available through Patient Access, Provider Access, and Payer-to-Payer APIs, which will increase the visibility of coverage criteria into machine-readable formats.
FDA's AI guidance does not address payer use
FDA's January 2025 draft guidance, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, addresses the use of AI in drug development and regulatory submissions, but it does not extend to payer use of AI in coverage decisions. Similarly, EMA's October 2024 Reflection Paper on AI in the Medicinal Product Lifecycle focuses on drug development and pharmacovigilance, not payer policy generation. This regulatory gap means that there is no federal oversight of AI-generated payer policies, and manufacturers must rely on existing appeal mechanisms and state-level regulations to challenge restrictive criteria.
Monitoring and response strategies for market-access teams
Strategy 1: Systematic payer-policy surveillance
Manufacturers should implement systematic monitoring of payer coverage policies, PA criteria, and formulary decisions for their products. This monitoring should cover:
- Major commercial payers and PBMs (UnitedHealth, CVS Caremark, Cigna Express Scripts, Aetna)
- Medicare Advantage plans
- State Medicaid fee-for-service and managed care formularies
- Specialty pharmacy PA criteria
The monitoring should flag any criterion that is narrower than the FDA label, including age restrictions, prior-therapy requirements, biomarker requirements, quantity limits, and comorbidity exclusions that are not supported by the label.
Strategy 2: Label-criterion gap analysis
For each payer policy, create a side-by-side comparison between the labeled indication, dosing, and patient population and the payer's coverage criteria. Document every gap where the payer criterion is more restrictive than the label. This analysis serves as the foundation for engagement with the payer's medical affairs or pharmacy affairs team.
Strategy 3: Evidence packages for criterion correction
When a payer policy is more restrictive than the label, prepare a targeted evidence package that includes:
- The relevant sections of the FDA-approved prescribing information
- Post-marketing data or real-world evidence supporting use in the excluded population
- Published clinical guidelines from professional societies that support broader use
- Cost-effectiveness or health-economic data if the restriction appears to be cost-driven
Submit the evidence package through the payer's formal policy-reconsideration process. For Medicaid programs, engage with state pharmacy and therapeutics committees during the formulary review cycle.
Strategy 4: State-level accumulator and PA reform legislation
Multiple states have enacted or are considering legislation to limit restrictive PA practices. The AMA's 2025 survey found that 93% of physicians report delays in patient care related to PA requirements. States including California, New York, Texas, and others have passed PA reform laws that require gold-carding (exemption from PA for providers with high approval rates), continuity-of-care requirements, and transparency in PA criteria. Manufacturers should track state-level PA reform legislation and engage with advocacy organizations pushing for criteria transparency.
Strategy 5: Engagement with AI policy development
As CMS's proposed rule moves toward finalization, manufacturers should engage with the regulatory process to advocate for transparency requirements in AI-generated PA decisions. The MACPAC analysis identifies "limited transparency" as a key challenge, and CMS's proposed API requirements are a step toward making PA criteria machine-readable and auditable. Manufacturers should comment on proposed rules, participate in CMS technical advisory panels, and support industry positions that require payers to disclose when AI is used to generate coverage criteria and to provide human review mechanisms.
The GLP-1 case study: where label restrictions are evolving
GLP-1 receptor agonists illustrate the dynamic tension between label and payer criteria. A May 2026 PMC analysis of affordable access to GLP-1 obesity medications noted that "coverage criteria narrower than the FDA label language can be criticized by clinicians and patients as being inappropriately restrictive" and that "using BMI thresholds to narrow use will become less feasible" as GLP-1 drugs obtain new indications for obstructive sleep apnea and cardiovascular risk reduction. As payers deploy AI tools to manage GLP-1 PA volume, the risk of restrictive criteria being encoded at scale increases. Manufacturer teams managing GLP-1 access should monitor payer criteria monthly, given the pace of label expansion and policy change.
Sources
- FDA. Enhancing Participation in Clinical Trials—Eligibility Criteria, Enrollment Practices, and Trial Designs. December 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents
- FDA. Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products. January 2025. https://www.fda.gov/regulatory-information/search-fda-guidance-documents/considerations-use-artificial-intelligence-support-regulatory-decision-making-drug-and-biological
- CMS. 2026 CMS Interoperability Standards and Prior Authorization for Drugs Proposed Rule. https://www.cms.gov/newsroom/fact-sheets/2026-cms-interoperability-standards-prior-authorization-drugs-proposed-rule
- MACPAC. Automation in the Prior Authorization Process. February 2025. https://www.macpac.gov/wp-content/uploads/2025/02/07_February-Slides_Automation-in-the-Prior.pdf
- Pharmaceutical Executive. Beyond Approval: Why Your Oncology Label Determines Commercial Success. 2025. http://www.pharmexec.com/view/beyond-approval-oncology-label-determines-commercial-success
- AJMC. Exploring the Complex, Ever-Evolving Landscape of Specialty Drug Coverage. https://www.ajmc.com/view/exploring-the-complex-ever-evolving-landscape-of-specialty-drug-coverage
- PMC. Affordable Access to GLP-1 Obesity Medications: Strategies to Guide Market Action and Policy Solutions in the US. May 2026. https://pmc.ncbi.nlm.nih.gov/articles/PMC12403326
- PMC. Commercial Coverage of Specialty Drugs, 2017–2021. March 2024. https://pmc.ncbi.nlm.nih.gov/articles/PMC10986192
- Tufts CEVR. Despite Criticisms of Accelerated Approval Pathway—Commercial Payers Defer to FDA. 2023. https://cevr.tuftsmedicalcenter.org/news/despite-criticisms-of-accelerated-approval-pathway-commercial-payers-defer-to-fda
- FDLI. Regulating the Use of AI in Drug Development: Legal Challenges and Compliance Strategies. July 2025. https://www.fdli.org/2025/07/regulating-the-use-of-ai-in-drug-development-legal-challenges-and-compliance-strategies
