Framework · FDA AI Lifecycle (readiness)
Readiness evidence for the AI-enabled devices you operate.
FDA's predetermined change control plan guidance is final as of December 2024; its lifecycle guidance for AI-enabled device software functions is still a draft from January 2025. Both address device sponsors. Meilynx evidences the deployer side: when the model behind a device function changed, runtime monitoring inputs, and a lifecycle record, with the sponsor's plan, data management, human factors, and post-market plan recorded as attestations. A readiness preset with no submission, classification, or clearance claim.
A change plan, data, human factors, post-market monitoring.
The PCCP guidance is final; the lifecycle guidance is a draft whose recommendations may change before it is finalized.
- A predetermined change control plan describing planned modifications, the modification protocol, and an impact assessment, so a sponsor can implement them without a new submission (final, December 2024).
- Data management: the sources, collection, annotation, representativeness, and independence of development and test data (draft, January 2025).
- Human factors: how users interact with the output, automation bias, and the information needed to interpret results (draft, January 2025).
- Post-market performance monitoring with drift and data-shift triggers, and lifecycle record-keeping (draft, January 2025).
Runtime expectations, to runtime evidence.
A specific Meilynx control for each expectation the proxy can substantiate, and the artifact it produces. Everything else is attested in the package, and the package says which is which.
FDA guidances → Meilynx controls (deployer readiness)
| Requirement | How Meilynx maps | Examination artifact |
|---|---|---|
Detect when the model behind a device function changed FDA PCCP (final, Dec 2024) · lifecycle guidance (draft) | The model registry records version changes for AI-assisted workflows. Whether an observed model is a device function and whether a change fell within the sponsor's plan are your determinations, recorded on the PCCP control. | Version-change evidence and PCCP attestation |
Feed the post-market monitoring plan FDA lifecycle guidance (draft) — post-market performance monitoring | Runtime findings and allow-list enforcement for the AI-traffic slice are one input to the monitoring plan; clinical performance monitoring is your program, attested. | Monitoring findings and post-market plan attestation |
Keep the lifecycle record FDA lifecycle guidance (draft) — record-keeping | The hash-chained AI-traffic record covers the deployment slice of total-product-lifecycle documentation; development and validation records are the sponsor's. | Hash-chained lifecycle record |
Record data management and human factors FDA lifecycle guidance (draft) — data management · human factors | The sponsor's data descriptions and the human-AI workflow design are attested per device function, with human-review records as supporting evidence. | Data management and human-factors attestations |
Detect when the model behind a device function changed
FDA PCCP (final, Dec 2024) · lifecycle guidance (draft)
Maps to · The model registry records version changes for AI-assisted workflows. Whether an observed model is a device function and whether a change fell within the sponsor's plan are your determinations, recorded on the PCCP control.
Examination artifact · Version-change evidence and PCCP attestation
Feed the post-market monitoring plan
FDA lifecycle guidance (draft) — post-market performance monitoring
Maps to · Runtime findings and allow-list enforcement for the AI-traffic slice are one input to the monitoring plan; clinical performance monitoring is your program, attested.
Examination artifact · Monitoring findings and post-market plan attestation
Keep the lifecycle record
FDA lifecycle guidance (draft) — record-keeping
Maps to · The hash-chained AI-traffic record covers the deployment slice of total-product-lifecycle documentation; development and validation records are the sponsor's.
Examination artifact · Hash-chained lifecycle record
Record data management and human factors
FDA lifecycle guidance (draft) — data management · human factors
Maps to · The sponsor's data descriptions and the human-AI workflow design are attested per device function, with human-review records as supporting evidence.
Examination artifact · Data management and human-factors attestations
What a deployer can show.
A readiness package that opens by saying it is one: the PCCP guidance final, the lifecycle guidance a draft, every section labelled with its evidence scope.
In the package
- Device version-change evidence with the PCCP attestation.
- Post-market monitoring inputs and the monitoring plan attestation.
- Data management and human-factors attestations.
- The lifecycle record for the deployment slice.
- Obligation timeline: PCCP final (December 2024), lifecycle guidance draft (January 2025), HIPAA pointer.
FDA AI Lifecycle and the proxy.
Is this an FDA submission or clearance claim?
No. The guidances address device sponsors; this preset is the deployer's readiness evidence and makes no marketing-submission, classification, or clearance claim. It is offered to healthcare organizations as optional, for those that operate AI-enabled device functions.
Why is the lifecycle guidance labelled draft everywhere?
Because it is. The January 2025 lifecycle guidance is a draft whose recommendations may change; every citation to it in the package carries that status so a reader can never mistake a recommendation for an obligation.
Which date is right for the PCCP guidance?
December 2024. The final guidance on predetermined change control plans for AI-enabled device software functions was issued then; an earlier internal reference to August 2025 was wrong and has been corrected everywhere.
See exactly what an examiner receives
Download a sample examination package: model inventory, control coverage, a governance policy snapshot, and a SHA-256 integrity hash.