An established global medical device manufacturer sought AlvaMed’s support in assessing a new web-based medical device software product being developed for a future FDA 510(k).
Generative AI had been used to accelerate portions of software development and supporting documentation. The result was a substantial initial software documentation package, but the manufacturer needed to understand whether that package was sufficient to support FDA expectations and what additional work would be required before submission.
The engagement also raised a newer regulatory question: How should the use of AI-assisted coding be addressed when AI helps develop the software but is not itself part of the medical device’s clinical function?
Challenge
The manufacturer had achieved significant development progress using AI-assisted tools, but several questions remained:
- How should AI-assisted software development be treated from an FDA and quality system perspective?
- Does use of generative AI for code development need to be disclosed in a 510(k)?
- Do FDA expectations for AI/ML model training apply when AI is used to generate code rather than clinical outputs?
Without clear regulatory guidance, internal teams were spending considerable time debating FDA expectations, creating delays in decision-making and uncertainty around commercialization timelines.
Approach
AlvaMed conducted a detailed regulatory and software quality assessment of the software documentation package against current FDA expectations and applicable software lifecycle principles. The engagement included:
- Reviewing the software development approach and AI-assisted development process.
- Assessing the completeness of the software lifecycle documentation against FDA expectations.
- Identifying documentation gaps required to support a compliant submission.
- Evaluating the regulatory implications of AI-assisted software development.
- Delivering a detailed gap assessment report outlining required documentation and recommended next steps.
- Presenting the findings to executive leadership to align technical and business stakeholders on the regulatory path forward.
The review identified several areas requiring further maturity, including software configuration management, code review and issue management, detailed software design, requirements/design/risk/test traceability, verification evidence, cybersecurity documentation and testing, labeling, and AI development governance.
A Critical Regulatory Distinction: AI-Assisted Development vs. an AI-Enabled Device
One of the most important conclusions from the assessment was that using AI to help develop medical device software is not the same as incorporating AI into the medical device function.
The product under review performed deterministic clinical calculations using predefined mathematical models. Generative AI had reportedly assisted with software development, but the AI tool was not incorporated into the marketed device and did not generate or influence clinical outputs.
Based on current FDA guidance and industry practice, AlvaMed concluded that the use of AI-assisted coding tools should not, by itself, represent a significant regulatory obstacle.
However, this did not eliminate the need for additional work.
The manufacturer still needed to demonstrate that the resulting software was developed under appropriate software lifecycle and quality system controls. This included evidence that the released code was understood and reviewed by qualified personnel, appropriately risk-managed, verified and validated, secured, configuration-controlled, and formally approved for release.
AlvaMed also recommended establishing internal governance for AI-assisted development, including expectations for approved tools, acceptable use, human review, verification, developer competency, confidentiality and PHI protection, and accountability for the final released software.
What About AI Training Data and Prompts?
The assessment also distinguished FDA expectations for AI/ML-enabled medical devices from the use of a general-purpose AI coding assistant.
When a trained AI/ML model is incorporated into a medical device function, FDA may evaluate topics such as model training and validation datasets, representativeness, bias, generalizability, model performance, and lifecycle management.
Those considerations were not directly applicable to this product because its clinical outputs did not depend on a trained AI/ML model.
Similarly, current FDA software guidance does not establish a specific requirement to submit AI prompts, development conversations, or the training methodology of a general-purpose coding model as part of a 510(k).
Instead, the regulatory evidence remains centered on the resulting medical device software and the controls supporting its development.
Results
AlvaMed helped the client translate regulatory uncertainty into a clear path toward 510(k) readiness and software quality compliance.
The engagement resulted in:
- A prioritized roadmap for completing FDA-ready software documentation.
- Early identification of software lifecycle, V&V, cybersecurity, traceability, and design control gaps.
- Clearer understanding of FDA considerations for AI-assisted software development.
- Recommendations for bringing AI-assisted code under appropriate review, verification, configuration management, and quality system controls.
- Alignment on the technical evidence and documentation needed to support a future 510(k).
AlvaMed also clarified an important distinction: using AI to develop software is different from incorporating AI into the medical device function. Here, AI assisted development but did not generate or influence clinical outputs. Current FDA guidance does not establish a specific affirmative disclosure requirement for use of an AI coding assistant; however, the resulting software must still meet applicable software lifecycle and design control expectations.
Key Services Provided
- FDA 510(k) regulatory strategy
- Software quality and lifecycle assessment
- Software documentation gap assessment
- FDA cybersecurity readiness assessment
- AI-assisted development regulatory assessment
- Design control and traceability review
- Executive regulatory advisory support
Key Takeaway
AI can accelerate medical device software development, but it does not replace FDA expectations for robust design controls and objective evidence. AlvaMed helped the client preserve those efficiencies while identifying the gaps and establishing a clear path toward 510(k) readiness.
