FDA Seeks Public Feedback on Regulatory Approach for Generative AI-Enabled Medical Devices

On August 18, 2026, the U.S. Food and Drug Administration (FDA) issued a discussion paper outlining its initial considerations for regulating generative artificial intelligence (GenAI)-enabled medical devices. The paper, developed by the Digital Health Center of Excellence (DHCoE) within the Center for Devices and Radiological Health (CDRH), is intended to engage stakeholders on the challenges associated with GenAI-enabled medical devices and ways to advance regulatory approaches for them. It addresses risk assessment, premarket evaluation, postmarket monitoring, and considerations specific to foundation models and agentic AI systems. The FDA is soliciting public comment through October 19, 2026.

This alert summarizes the key elements of the discussion paper and identifies practical steps that device manufacturers, developers, and other stakeholders should consider in responding to the FDA’s request for feedback.

Key Takeaways

• The discussion paper is a framework for dialogue, not draft guidance or policy. It may signal FDA thinking, but it sets no current regulatory expectations and creates an important window for industry input.
• The FDA proposes a two-axis risk framework that would calibrate regulatory expectations based on both the clinical significance and autonomy level of a GenAI-enabled device.
• A novel “competency assessment” approach to premarket evaluation, modeled conceptually on physician training, would consist of non-clinical benchmarking and clinical confirmation.
• Postmarket monitoring may need to address unique challenges of non-deterministic systems, including monitoring not only the manufacturer’s GenAI model but the underlying foundation model.
• Comments are due by October 19, 2026, under docket FDA-2026-N-7874 on Regulations.gov.

Background

The FDA released a discussion paper titled, “Considerations for the Regulation of Generative AI-Enabled Medical Devices,” developed by the DHCoE, which supports the FDA’s Public Health Pillar on “Innovation and Global Leadership.” The effort aligns with the current Administration’s priority to harness AI to accelerate delivery of innovative medical products.

The FDA defines a “GenAI-enabled device” as a device (as defined in section 201(h) of the FD&C Act) in which GenAI methods or models are integral to the device’s output or functionality. The paper emphasizes that these devices hold transformative promise but may introduce unique risks compared to traditional software and AI-enabled medical devices, particularly given their non-deterministic nature and potential for continuous adjustment after deployment.

Importantly, this paper is issued solely for discussion purposes. It does not represent draft or final guidance, is not intended to propose or implement policy changes regarding how CDRH intends to regulate GenAI-enabled devices, and is not intended to communicate CDRH’s proposed or final regulatory expectations, including its expectations for supporting evidence in future marketing submissions. Rather, it seeks early input from outside the Agency and aims to advance a broader discussion. 

Substance of the Discussion Paper

The discussion paper is organized around four topic areas, each accompanied by questions posed by FDA: (a) considerations for the assessment of risk for GenAI-enabled medical devices; (b) considerations for premarket evaluation of GenAI-enabled medical devices; (c) postmarket monitoring for GenAI-enabled medical devices; and (d) other topics relevant to the regulation of GenAI-enabled medical devices. The sections below summarize these areas, including the paper’s discussion of foundation models and agentic AI systems.

Two-Axis Risk Framework

The discussion paper outlines a possible two-axis framework for risk assessment that could inform the level of regulatory scrutiny applied to GenAI-enabled devices. While the paper does not specify the exact axes, it suggests that risk-based calibration would account for factors such as:

• The clinical significance of the device’s output or intended use (e.g., diagnostic, therapeutic, informational); and
• The level of autonomy of the GenAI system (e.g., clinician-in-the-loop versus fully autonomous operation).

This risk framework builds on the FDA’s longstanding total product life cycle (TPLC) approach to device oversight, under which regulatory requirements are proportionate to the intended use and technological characteristics of the device.

Premarket Evaluation: Competency Assessment

The paper introduces a potential premarket evaluation approach built on the concept of “competency assessment,” inspired at a high level by how physicians are trained and evaluated. This approach would consist of two components:

1. Non-clinical device benchmarking: standardized performance testing against defined benchmarks to evaluate the device’s baseline capabilities, accuracy, and reliability.
2. Clinical confirmation: real-world or simulated clinical evaluation to confirm that the GenAI-enabled device performs as intended before reaching patients.

This two-stage approach reflects the FDA’s recognition that traditional software validation methods may not fully capture the performance characteristics of non-deterministic GenAI systems.

Risk-Proportionate Postmarket Monitoring

Recognizing that GenAI devices are non-deterministic and may undergo continuous adjustment after deployment based on localized live data, user interactions, and changing conditions, the discussion paper describes several potential approaches to postmarket monitoring. Key considerations include:

• Monitoring not only the manufacturer’s own GenAI model but also the underlying foundation model for performance drift, hallucinations, or unexpected outputs;
• Calibrating postmarket requirements to the level of risk identified under the two-axis framework;
• Addressing the complexities introduced by continuous learning and adaptation in deployed devices; and
• Tracking adverse events and other safety signals unique to generative AI outputs.

The FDA poses targeted questions regarding how to design a postmarket framework that is scientifically rigorous and proportionate to the novel capabilities of GenAI-enabled devices.

Foundation Models and Agentic AI Systems

The discussion paper also addresses considerations specific to foundation models, large-scale pre-trained models that may underlie multiple device applications, and agentic AI systems that can take autonomous or semi-autonomous actions.

The FDA recognizes that manufacturers of GenAI-enabled devices may rely on third-party foundation models, raising questions about transparency, accountability, and the allocation of regulatory responsibility across the supply chain. The paper invites comment on how premarket and postmarket expectations should apply when the device manufacturer does not control the underlying foundation model.

Public Comment Process

The FDA is seeking feedback from device manufacturers, clinicians, researchers, the public, and other interested parties. For each topic area addressed in the discussion paper, the FDA poses targeted questions designed to inform the development of a regulatory framework that prioritizes patient safety while supporting innovation.

Stakeholders are encouraged to provide substantive responses to the FDA’s specific questions as well as broader comments on the overall regulatory approach described in the paper.

Recommended Actions

Although this discussion paper does not establish current regulatory expectations, it provides significant insight into the FDA’s thinking and creates a meaningful opportunity for stakeholders to shape the regulatory landscape. Companies developing or commercializing GenAI-enabled medical devices should consider the following steps:

Submit Comments. The October 19, 2026 deadline provides a limited window to influence the FDA’s approach. Manufacturers should prepare substantive responses to the FDA’s specific questions, particularly regarding the competency-assessment concept, postmarket monitoring considerations, and the treatment of foundation models.

Assess Product Portfolios. Companies should evaluate which products in their pipeline or on the market may qualify as “GenAI-enabled devices” under the FDA’s proposed definition and assess how the two-axis risk framework might apply.

Review Foundation Model Relationships. Manufacturers relying on third-party foundation models should evaluate contractual arrangements, transparency obligations, and the ability to monitor model performance, as the FDA’s approach may impose requirements that extend beyond the device manufacturer’s own model.

Develop Validation Frameworks. Consider how existing testing and validation programs map to the proposed competency-assessment approach (non-clinical benchmarking and clinical confirmation). Identify gaps that may need to be addressed if this framework is adopted.

Monitor Developments. This discussion paper is the beginning of a process. Companies should track FDA workshops, advisory committee meetings, and subsequent guidance documents as the regulatory framework takes shape.

For More Information

For questions about the FDA’s discussion paper, the public comment process, or how these developments may affect your regulatory strategy, please contact the authors listed herein.

ATTORNEY ADVERTISING

This client alert is provided for informational purposes only and does not constitute legal advice. The information contained herein should not be relied upon or regarded as a substitute for specific legal advice from counsel. No attorney-client relationship is created by the distribution or receipt of this alert. Recipients should consult their own legal advisors regarding the application of the law to their specific facts and circumstances. Prior results do not guarantee a similar outcome. All rights reserved.