Aleksandar Josipovic

# Public Comment ? Docket No. FDA-2026-N-7874

**Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper and Request for Feedback**

**Submitted by Dr Aleksandar Josipovic, DBA, in an individual capacity.**

The observations below reflect my independent professional perspective and do not represent the views of any company, client or organisation.

My background is in international business strategy, communication and the commercial adoption of innovation, including experience in healthcare-related industries. My comments focus on the relationship between AI-generated information, user understanding and responsible technology adoption. They do not address the performance of any specific product or manufacturer.

### 1. Communication and the Directiveness of AI Outputs (Q2)

I welcome CDRH’s consideration of directiveness as a continuum rather than a binary distinction.

The same information can influence decisions differently depending on its wording, specificity, degree of personalization and expression of certainty.

I suggest that evaluation approaches consider a small set of observable communication characteristics, including whether an output presents information, implies a recommendation or explicitly directs an action.

Assessing these characteristics through realistic user scenarios could help clarify how outputs behave in practice, alongside the device’s intended use and risk profile.

### 2. Patient Understanding and Appropriate Reliance (Q3)

Patient-facing generative AI has the potential to improve access to information and support more informed participation in healthcare decisions.

An important consideration is whether intended users understand both the information provided and its limitations.

Where appropriate to a device’s intended use and risk profile, comprehension testing with representative users could help assess whether people correctly interpret uncertainty, recognize the limits of the information and understand when professional clinical input may be needed.

Such evaluation should support patient autonomy without assuming that every patient has the same level of health literacy or familiarity with medical terminology.

### 3. Communication Across Languages and User Groups (Q9)

As generative AI-enabled devices may support users with different linguistic and educational backgrounds, communication performance deserves attention within assessments of generalizability.

For languages within a device’s intended use, relevant evaluation could consider whether important information, expressions of uncertainty and safety-related guidance remain understandable and consistent.

A risk-proportionate approach could help distinguish communication differences that are acceptable from those that may materially affect user understanding or appropriate reliance on an output.

### 4. Practical Feedback and Postmarket Monitoring (Q21)

Clinicians and healthcare institutions may be well positioned to identify communication difficulties or unexpected outputs during real-world use.

Clear and accessible pathways for reporting such observations could contribute useful information to postmarket monitoring.

I suggest considering practical reporting mechanisms that allow relevant concerns to be documented and communicated to the appropriate manufacturer or responsible party, with roles and escalation processes defined in accordance with applicable requirements.

This could support effective communication among stakeholders without transferring or diluting established regulatory responsibilities.

### 5. Communication and Management of Foundation Model Changes (Q24)

Where a device relies on a third-party foundation model, changes to that model may affect outputs experienced by end users.

Depending on the device’s intended use, risk profile and technical architecture, useful safeguards could include change notifications where feasible, version tracking, targeted reassessment of relevant outputs and controlled implementation procedures.

These approaches may help manufacturers identify meaningful changes while allowing appropriate flexibility for technological development.

### Closing Observation

Responsible innovation depends not only on technological capability but also on how people understand and use the information that technology provides.

Clear communication, proportionate evaluation and defined responsibilities can help support user confidence and informed adoption while preserving opportunities for beneficial innovation.

I appreciate the opportunity to contribute to CDRH’s discussion.

Dr Aleksandar Josipovic, DBA

*Submitted in an individual capacity.*