When does an AI chatbot become a medical device, and how do you regulate technology that can change the way it responds to patients?

This week on Trying to Keep Up, Allison Komiyama and Michael Nilo are joined by Aubrey Shick, former senior digital health advisor at the FDA’s Digital Health Center of Excellence and founder of Launch and Logic, for a deep dive into generative AI, digital health policy, and how the FDA is approaching technologies that don't always fit neatly into traditional regulatory models.

Aubrey brings a unique perspective to the conversation. Before joining the FDA, she spent more than a decade working in product development and user experience. At the FDA, she helped develop and interpret digital health policy, including work involving generative AI and human-centered AI transparency.

The group looks at RecoveryAI's recent Breakthrough Device Designation for a physician-prescribed, AI-powered assistant designed to support patients recovering from joint replacement. The designation offers an early glimpse into how the FDA may evaluate patient-facing generative AI within existing medical device frameworks.

The conversation also explores the FDA's TEMPO pilot and what its implementation could tell the industry about real-world evidence, enforcement discretion, and emerging digital health policy. Aubrey explains why policy often develops through individual submissions before it appears in formal guidance and why companies working at the edge of existing policy should consider involving the Digital Health Center of Excellence early.

The team also examines the FDA's own growing use of AI. With Elsa reportedly used voluntarily by more than 70% of agency staff, they discuss transparency, data security, model dependency, and whether sponsors should know when AI contributes to a product review.

Additional topics include:

• The FDA's new unified Adverse Event Monitoring System

• Human-centered transparency for patient-facing AI

• The risks and opportunities of generative AI in mental health

• What sponsors should know about TEMPO applications

• How intended use can change a device's regulatory classification

• A new AI-enabled breast cancer imaging PMA

• The FDA's request for feedback on 510(k) submission burden

• CDRH early alerts and device reclassifications

• Digital health resources that innovators may be overlooking

• Why FDA databases provide transparency, but not always discoverability

Plus, Aubrey explains why digital health sometimes comes down to one fundamental question: “Is this a device, or are we all just having a very expensive misunderstanding?”