The US Food and Drug Administration (FDA) began the new year by publishing draft guidance titled Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations (the Draft AI Guidance).¹ This guidance offers detailed recommendations to support the development and marketing of safe and effective AI-enabled devices, including by identifying the information and documentation the FDA requests for marketing submissions² for devices that include AI-enabled software functions.³ The Draft AI Guidance, which is voluntary, also includes guidance on the design, development, deployment, and maintenance of AI-enabled devices.

The Draft AI Guidance was issued in the final weeks of the Biden administration and is based on the agency’s authorization of more than 1,000 AI-enabled devices through established premarket pathways. Although comments on the draft guidance will be accepted through April 7, 2025,⁴ it is unclear how this guidance will fit into the larger Trump administration’s approach to AI. In one of his first actions, President Trump issued an Executive Order rescinding the Biden administration’s sweeping executive order regulating AI⁵ and later issued one of his own, marking a potentially significant shift in federal agency oversight of, and perspective on, AI. The Draft AI Guidance, however, is of a more practical and administrative nature.

Below we outline key aspects of the Draft AI Guidance and its recommendations, as well as key takeaways for businesses across the healthcare sector.

Guidance on Premarket Submissions

The Draft AI Guidance provides a detailed explanation as to what information the agency recommends that sponsors include in an AI-enabled device marketing submission, where such information should be included, and why. The FDA recommends, among other things, as follows:

• Device Description: Sponsors should provide a detailed description of the AI-enabled device, including a statement that AI is used in the device, its intended use, inputs, outputs, operating environment, and a description of the degree of automation the device provides in comparison to the current standard of care. The FDA is especially interested in understanding how AI contributes to the device’s functionality, including how its outputs will be integrated into the clinical workflow.
• User Interface: Sponsors should include a clear explanation of the software architecture, with a focus on the design of the user interface, operational workflows, and how the system integrates into existing healthcare environments. The requirements for both device and software descriptions align closely with current practices for medical devices incorporating software functions.
• Labeling: Sponsors should include user-friendly instructions and clear explanations of AI functionality to ensure all stakeholders, including clinicians, technicians, and patients, can understand and safely use the device.
• Risk Assessments: Sponsors should provide a comprehensive risk management file, including a risk assessment plan, identification of potential hazards, and strategies for identifying and mitigating risks throughout the device lifecycle. In this regard, the Draft AI Guidance is consistent with prior guidance on the Content of Premarket Submissions for Device Software Functions.
• Data Management: Sponsors should include clear documentation of how data is collected, annotated, validated, and managed. To mitigate AI bias and ensure the device performs consistently across demographic groups and clinical settings, the FDA recommends using high quality, diverse, and representative datasets.
• Model Description and Development: Sponsors should include documentation of the AI model’s design beyond device and software descriptions. This includes providing details on the model’s architecture, development process, and the datasets used for training and validation.
• Performance Validation: Sponsors should perform rigorous testing and provide documentation to demonstrate the device’s safety and efficacy. This includes providing performance metrics (such as for sensitivity, specificity, and accuracy), along with subgroup analyses to ensure equitable outcomes.
• Device Performance Monitoring: Sponsors should include information regarding their performance monitoring plans. The FDA highlights that AI-enabled devices are especially vulnerable to performance changes or degradation over time due to the heightened sensitivity of AI models to variations in data inputs. As such, the Draft AI Guidance encourages robust post-market monitoring systems to identify and address issues such as data drift and maintain performance over time.
• Cybersecurity: Sponsors should identify any unique considerations related to AI cybersecurity, and also provide a “Security Use Case View” addressing AI-enabled considerations for the device, a description of controls implemented to address data vulnerability and prevent data leaks, and an explanation of how cybersecurity testing addresses the risks associated with the model. The Draft AI Guidance is consistent with the submission documentation recommended in the 2023 Premarket Cybersecurity Guidance regarding the cybersecurity controls and security risk management relevant to the AI components or features.

The FDA also makes available resources to assist in preparing marketing submissions for AI-enabled devices. For instance, the Draft AI Guidance includes examples of submission language and an appendix outlining recommendations for the contents and formatting of a “model card,” designed to support clear and consistent communications about the AI-enabled device to interested parties. While not required, the agency emphasizes that use of a “model card” is a potential means to consistently summarize the key aspects of AI- enabled devices and can be used to concisely describe their characteristics, performance, and limitations. 

Reinforcing the Total Product Lifecycle (TPLC) Approach

More generally, building on its long-standing commitment to a TPLC approach to oversight of medical devices, the FDA’s Draft AI Guidance references various regulatory obligations (including with regard to cybersecurity) and cross-references other agency guidance in order to assist manufacturers with applying these regulatory authorities and earlier FDA recommendations to AI-enabled devices. It also provides recommendations and resources in support of the design, development, validation, marketing, and ongoing oversight of AI-enabled devices.

Key Takeaways

• Helpful Guideposts: The Draft AI Guidance is intended to provide developers of AI-assisted devices with a better understanding of the information that the agency will find helpful in its review of products and how such information should be presented. While it is unclear in what format the draft guidance will be finalized, it nevertheless provides helpful insight into the agency’s current perspective on what information should be submitted in support of an application, why, and in what context.
• Comprehensive Lifecycle Management Should Be the Norm: The FDA emphasizes a TPLC approach that requires robust risk management, ongoing performance monitoring, and mechanisms to address algorithm updates and data drift. The guidance provides helpful insights into the FDA’s view of AI-enabled devices and strategies to address transparency, bias and other challenges arising in this context.
• Agency Engagement and Documentation: The draft guidance repeatedly encourages early, proactive engagement with the FDA during development of AI-enabled devices, particularly when new and emerging technology is used in the development and design of the device, or when novel methods are used during the validation of a device. Organizations should strategically consider taking advantage of this opportunity to collaborate with the agency.
• Responsible AI Governance: Organizations looking to engage with the FDA relating to AI-powered devices should ensure they have robust AI governance practices in place. While the Draft AI Guidance is most relevant to developers of AI-enabled devices, other health care organizations could benefit from examining the FDA’s recommended approach as many of the recommendations in the Draft AI Guidance have potential application to AI model development, performance validation, and risk mitigation in other contexts as well.

If you would like to learn more about the Draft AI Guidance or about Dentons’ AI and Healthcare teams, please reach out to one of the co-authors listed in this guidance.

1. While the FDA has issued numerous guidance documents over the years, the Draft AI Guidance, if finalized, “would be the first guidance to provide comprehensive recommendations for AI-enabled devices throughout the total product lifecycle, providing developers an accessible set of considerations that tie together design, development, maintenance and documentation recommendations to help ensure safety and effectiveness of AI-enabled devices.” FDA Draft Guidance for Industry: Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, p. 2, (Jan. 7, 2025).↩
2. For purposes of the Draft AI Guidance, the term “marketing submission” refers to a premarket notification (510(k)) submission, a De Novo classification request, a Premarket Approval (PMA) application, a Humanitarian Device Exemption (HDE), or a Biologics License Application (BLA).↩
3. In the Draft AI Guidance, the term “AI-enabled device” is a device that includes one or more AI-enabled device software functions (AI-DSFs).  An AI-DSF is a device software function that implements one or more “AI models” to achieve its intended purpose.  An AI model is a mathematical construct that generates an inference or prediction based on new input data.↩
4. In addition to general comments, the FDA is specifically requesting public comment on the draft guidance’s “alignment with the AI lifecycle; the adequacy of the recommendations to address concerns that may be raised by emerging technology such as generative AI; the approach to performance monitoring (including use of a performance monitoring plan as a means of risk mitigation for AI-enabled devices); the type of information about AI-enabled devices that should be conveyed to users and the most appropriate approach to deliver that information.” FDA News Release, "FDA issues Comprehensive Draft Guidance for Developers of Artificial Intelligence-Enabled Medical Devices” (Jan. 6, 2025).↩
5. See, President Biden’s Executive Order 14110, Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence (Oct. 30, 2023).↩