Should Epic open Cosmos data to third-party AI developers or keep it reserved for Epic and its customers?

Introduction

Cosmos is Epic's de-identified dataset contributed by participating Epic customers, covering about 320 million unique patients by 2026. Today it is used mainly by clinicians and researchers at Epic organisations, through tools like SlicerDicer, Look-Alikes and Best Care Choices, and powers Epic's own predictive models. More than 220 research papers have been published using Cosmos. The prompt: should Epic open Cosmos to third-party AI developers, such as startups and pharmaceutical companies, to train and validate models, or keep it reserved for Epic and its customers? This trade-off pits ecosystem growth and scientific impact against privacy, customer trust and Epic's competitive advantage.

Step 1

Clarifying Questions (3 minutes)

• Context: What kind of access is being considered?

Why it matters: Raw data export is very different from running approved analyses inside a secure environment. Expected answer: Access inside a controlled environment, with no data leaving it. Impact on approach: Privacy risk is lower but not zero.

• Business Context: Would Epic charge for access?

Expected answer: Possibly, but Cosmos was built on customer contributions, so any commercial use must be acceptable to customers. Impact on approach: Customer consent and governance are central.

• User Impact: Who are the third parties?

Expected answer: AI startups, academic groups outside Epic customers, and life sciences companies. Impact on approach: Each has different trust levels and uses.

• Technical: Is the platform ready for external users?

Expected answer: Partially; it would need new access controls, auditing and compute isolation.

• Resource: What else would the Cosmos team build otherwise?

Expected answer: More Epic AI features like Cosmos predictions in Agent Factory.