Microsoft and Epic Systems are embedding generative AI from Microsoft's Azure OpenAI Service, including GPT-4, straight into Epic's electronic health record software. The initiative targets two tasks: drafting replies to messages from patients and condensing medical histories. The goal is to ease the administrative load that fuels burnout and staffing gaps. This move extends a longstanding cloud deal between the two companies, shifting the relationship from infrastructure hosting to application-level AI that clinicians interact with directly.
The integration runs on Azure OpenAI Service, the same platform behind Microsoft's Copilot products. Epic weaves the models into its EHR workflows so a physician reviewing a chart can generate a draft summary or a portal reply without switching screens. As of April 2023, neither company named the health systems testing the technology, nor did they set a general-availability date. The features were described as under development and in early evaluation.
Epic holds the largest share of the U.S. hospital EHR market. Any change to its system touches hundreds of hospitals and thousands of physicians. The alliance signals that Microsoft sees healthcare AI as a deployment challenge rather than a research exercise: the models exist, and the bottleneck is fitting them into regulated clinical routines.
Models and Cloud Services
The collaboration draws on Azure OpenAI Service, giving Epic access to GPT-4 and other large language models running inside Microsoft's cloud. Microsoft began selling this commercial service in January 2023. It lets enterprise customers use OpenAI's models within the Azure boundary, under data-handling policies that differ from OpenAI's public API.
Epic does not host the models on its own servers. Instead, patient data travels to Azure OpenAI Service for inference. The models are not fine-tuned on patient records in the usual sense; Microsoft and Epic state that prompts and outputs stay inside the Azure trust boundary and are never used to retrain the base models. That separation matters for HIPAA compliance, which requires covered entities to sign business associate agreements with any vendor touching protected health information. Microsoft offers such agreements for Azure OpenAI Service. Epic, as a software supplier, also signs them with its customers.
Choosing GPT-4 over earlier releases such as GPT-3.5 is significant for medical settings. GPT-4 produces fewer hallucinations on healthcare benchmarks, though neither company claims the model is safe for autonomous decision-making. The stated uses remain assistive: drafting and summarizing, not diagnosing or treating.
Clinical Use Cases: Drafting and Summarizing
Message drafting
The first use case generates draft replies to messages that arrive through Epic's patient portal. The AI reads the person's question along with the relevant slice of the medical record and proposes a response. The clinician reviews, edits, and sends the final message.
Record summarization
The second use case pulls structured and unstructured data from the EHR to produce a digest of a person's medical history. That summary can support care coordination, prior authorization, and shift handoffs.
What the AI does not do
Neither use case involves making a diagnosis, ordering a test, or prescribing a drug. Both are documentation and communication chores that eat up a large share of a clinician's day. A 2022 study in the Annals of Internal Medicine found that physicians spent nearly two hours on EHR and desk work for every hour of direct care. Narrowing that ratio is the stated objective.
Adoption unknowns
The collaboration does not specify whether the AI will appear as an opt-in feature or be switched on by default. It also sidesteps the question of liability when a clinician misses an error in a draft. Those issues fall to Epic's customer contracts and the malpractice frameworks of individual health systems.
Regulatory Compliance and Data Privacy
HIPAA and infrastructure
Healthcare AI operates inside a thicket of regulation. HIPAA dictates how protected health information must be stored, transmitted, and processed. The alliance addresses this in two ways. First, data sent to Azure OpenAI Service remains inside Microsoft's Azure fabric, which carries HIPAA and HITRUST certifications. Second, Epic and Microsoft maintain a business associate agreement covering the use of Azure OpenAI Service for EHR data.
FDA classification
The tools are not classified as clinical decision support software under FDA rules. The agency has neither cleared nor approved GPT-4 for medical use. Microsoft and Epic describe the features as administrative aids, not decision-support products. That label carries weight because FDA-cleared software must meet stricter validation and monitoring requirements. If the tools later evolve to suggest diagnoses or treatments, they would face a different regulatory pathway.
Consent gaps
Patient consent remains an open question. Drafting a message reply requires the AI to access the person's medical record. Health systems deploying the feature will need to decide whether to notify individuals or obtain opt-in consent. As of April 2023, neither Microsoft nor Epic had publicly addressed that step.
How This Differs from the Prior Cloud Agreement
Microsoft and Epic have maintained a strategic alliance since at least 2019, when Epic agreed to run its EHR platform on Azure. That deal centered on infrastructure: Epic shifted its cloud hosting to Microsoft, giving hospitals scalability and compliance certifications without changing how clinicians interacted with the system day to day.
The 2023 generative AI initiative belongs in a different category. It embeds models into care workflows rather than simply hosting an application. Instead of moving data to the cloud, the new effort pulls model inference into the EHR interface. The earlier cloud arrangement made the technology stack cheaper and more secure. The generative AI push aims to reshape the speed and format of clinical communication.
Neither company disclosed financial terms for the expanded relationship. The earlier cloud deal was a multiyear agreement of undisclosed value. The AI initiative appears to extend that commercial arrangement rather than standing as a separate contract.
Executive Leadership and Announcement
Who spoke
The initiative was announced jointly by Microsoft and Epic. Microsoft's healthcare leadership includes executives who oversee its cloud and AI business in the health sector. Epic's side includes its CEO and senior product leaders. Specific titles were absent from the announcement materials.
Burnout narrative
Both companies framed the effort as an answer to workforce burnout and staffing shortages. The American Medical Association reported that burnout rates among physicians topped 50 percent in 2022. Cutting documentation time stands as one of the few interventions with measurable impact on clinician satisfaction. The public narrative ties directly to that data point.
Missing voices
The announcement included no quotes from health system executives who had tested the tools. It also named no hospital or clinic that had agreed to pilot the features. The absence of named early adopters suggests the integration was still in early development as of April 2023.
Deployment Status and Open Questions
Current state
As of April 2023, the generative AI features were in development and early testing. Neither company announced a date for pilot launch or general availability. The features were not yet broadly available in Epic's system. Health systems wanting to participate would need to work directly with Epic and Microsoft on a pilot basis.
Unanswered pricing and configuration questions
Several gaps remain. The alliance did not disclose whether the AI tools would be priced separately or bundled into existing Epic and Microsoft subscriptions. It did not specify whether health systems could choose which Azure OpenAI models to use or whether the tools would function with non-Microsoft cloud providers.
Sensitive data and special populations
The announcement also did not address how the AI would handle pediatric records, mental health information, or other sensitive data categories that carry extra privacy protections under HIPAA and state law.
Outlook
Key Facts
- Parties: Microsoft and Epic Systems
- AI service: Azure OpenAI Service, including GPT-4
- Target use cases: Drafting patient message responses, summarizing medical records
- Regulatory posture: HIPAA business associate agreement in place; not FDA-cleared
- Deployment status (as of April 2023): Development and early testing; no GA date announced
- Basis: Extension of existing cloud partnership (2019)
Frequently Asked Questions
Is GPT-4 being used to diagnose patients?
No. The stated use cases are drafting message responses and summarizing records. Neither Microsoft nor Epic has claimed the model is safe for autonomous diagnosis or treatment.
Does the integration comply with HIPAA?
Microsoft offers HIPAA business associate agreements for Azure OpenAI Service, and Epic has a BAA in place with Microsoft. Data is processed within Azure's infrastructure.
When will the features be available to hospitals?
No timeline was announced as of April 2023. The features were in development and early testing.
Which health systems are piloting the technology?
No health systems were named in the announcement.
