The Cat Is Out of the Bag: AI's Impact on Specialty Pharmacy Quality

Artificial intelligence (AI) has moved well past novelty in health care, and it is now deeply embedded in the workflows that specialty pharmacy professionals navigate daily. From prior authorization and member services to clinical documentation and drug safety surveillance, AI is being used. At the AXS2026 summit, a panel of 4 experts gathered to assess both the opportunity and the risk that comes with that reality as well as the importance of trust.

"The technology has progressed, but the trust is still chasing it," said Shawn Griffin, MD, president and CEO of Utilization Review Accreditation Commission (URAC). "As this comes into care, how can we protect patients and providers and pharmacists, so that the tools are worthy of their trust?"

Griffin highlighted the importance of this conversation as "most hospital systems have more regulation and oversight on the hospital cafeteria than on the AI program."

AI's Expanding Footprint in Pharmacy

Mitzi Wasik, PharmD, MBA, senior vice president for practice, strategy, and innovation at the Academy of Managed Care Pharmacy, said that prior authorizations are the clearest and most current example of AI use in the pharmacy field as payers and pharmacy benefit managers seek to reduce administrative friction.

"What we're seeing is how AI is impacting the member experience," Wasik said, noting that AI has dramatically shortened call wait times and improved the availability of real-time data for customer service representatives. However, according to a 2024 American Medical Association survey of 1000 practicing physicians, 61% expressed concern that AI tools used by health plans are increasing prior authorization denial rates.¹

Wasik also described a shift from ambient AI usage with tools that quietly assist in the background, toward agentic AI with systems that can execute complex, multi-step workflows autonomously while “allowing the human to stay in the loop for the harder cases," she explained. "Before, you had to work through 9 simpler cases to get to the 1 complicated one. You lost time for the 1 that you could have been spending on."

She also acknowledged that the regulatory landscape is struggling to keep pace. "We have some standards, but even our own federal government has not given us any true guardrails," she said. "You're really seeing all of the states kind of trying to jump in."

This fragmentation of state legislation is particularly burdensome for national payers. Wasik highlighted Utah's AI innovation sandbox as an example of how far and fast things are moving. She described an AI program operating within Utah's regulatory sandbox that uses AI to prescribe refills for a list of 150 drugs without pharmacist involvement. "Will AI be able to prescribe before pharmacists can?" she asked.

A Hidden Risk in Drug Safety

Heather Rubino, PhD, MS, head of safety surveillance research at Pfizer, brought a safety perspective. Her work involves postauthorization drug safety studies that measure what happens to patients and medicines once they leave the clinical trial and enter the real world. AI, she argued, is quietly reshaping the very data on which that measurement depends.

"There is no standardized, centralized way to track all of those pieces of information," Rubino said, describing the layered role AI now plays in a single clinical encounter from ambient recording tools, noise suppression algorithms, transcription engines, and AI-assisted note generation all contribute to the medical record that pharmacovigilance researchers ultimately rely on. "There's no way today that we can go back and track how many times and how many different models AI has already influenced the medical record."

This matters enormously, she explained, because if AI systematically over- or underrepresents certain clinical events in documentation, safety signals can become distorted. False results could mean a drug is pulled from the market or that who need it may not know it is safe. "Patient safety is as much protecting patients from drugs as it is protecting patient access to drugs," Rubino said.

Her concern is echoed in the literature with a 2025 review noting that AI models face significant implementation challenges including risks of algorithmic bias and inconsistent performance and emphasized that transitioning AI from experimental use to routine, scalable practice requires robust governance and continuous human oversight.²

"The cat is out of the bag," Rubino acknowledged, "and it's moving. There is no slowing that down."

Building the Guardrails: URAC's AI Accreditation

Heather Bonome, PharmD, director of pharmacy at URAC, explained why the organization felt compelled to act when federal regulation stalled. URAC, which has been accrediting health care organizations for 35 years, launched its Health Care AI Accreditation Program in September 2025, marking the first of its kind.

"There's a real need for national, flexible, adaptable standards as to how to manage AI in health care," Bonome said. "And that lends itself to accreditation really well."

The program is modular and covers two distinct tracks. A developer module addresses how AI systems are designed, trained, tested, and monitored. A user module focuses on how health care organizations implement and govern the AI tools they deploy. Many organizations applying for accreditation are seeking both. Critically, Bonome emphasized, the program does not require organizations to share source code or trade secrets.

"This is not a certification of your technical code," she said. "This is an accreditation of the principles and the quality principles employed as you're building, deploying, and maintaining your systems."

Disclosures for who needs to know when AI is involved in a clinical or administrative decision, and at what level, are a key element of the framework. Postdeployment monitoring requirements address the risk of model drift and false findings.

Bonome noted that the standards will likely be revised within a year or two as URAC gains real-world insight from organizations going through the review process. "We're going to have our first organizations achieve AI accreditation in the next month or two," she said. "I think that's really exciting."

URAC's standards and a related white paper are available free of charge on the organization's website.

Advice From the Panel

The session's final segment turned toward practical guidance. Wasik urged every attendee to examine whether their organization actually has an AI governance policy and to recognize that it cannot simply be an extension of the existing IT policy.

Rubino directed her advice toward engagement with the policy process itself. "The rules are being written by folks, some of [whom] know very little about the technology and don't know where it's going and don't know how it plays out on the ground—but you do," she told the audience. "Find out when that policy is next up for renewal and put in suggestions. Let's stop driving the car by looking in the rearview mirror and [instead] start looking through the windshield."

Griffin closed the session by revisiting the importance of trust in this conversation. "Think about any AI tool you're going to use in the delivery of care as though it has to go through credentialing to earn your trust, so that you can share it with the people that you care for in your communities."

REFERENCES

1. American Medical Association. AMA prior authorization physician survey. AMA; 2024. Accessed April 28, 2026. https://www.ama-assn.org/system/files/prior-authorization-survey.pdf

2. Nagar A, Gobburu J, Chakravarty A. Artificial intelligence in pharmacovigilance: advancing drug safety monitoring and regulatory integration. Ther Adv Drug Saf. 2025. doi:10.1177/20420986251361435