FDA's 2027 AI playbook pairs lifecycle and change-control rules

When the FDA's device center published its FY 2027 workplan on 1 October, it did something that AI medical-device makers have been waiting on since 2025: it scheduled finalization of the January 2025 AI lifecycle draft alongside the change-control framework that lets those models update after clearance.

CDRH's FY 2027 guidance agenda places the January 2025 draft on AI-Enabled Device Software Functions at the top of its A-list of priorities, alongside finalization of guidance on pre-determined change control plans, or PCCPs. Together, those two documents define the operating license for adaptive AI in clinic: lifecycle management for the model, and a change-control path for what happens when the model is retrained.

A PCCP is the mechanism that lets a manufacturer describe, in advance, the kinds of changes it plans to make to a device after it is on the market, and how it will validate those changes, without filing a new clearance each time. For adaptive AI, which is built to improve as it sees more data, that distinction is the whole game. Without a PCCP, every meaningful retraining can trigger a new submission, which is not a path any learning-health system can run on.

The 1 October workplan puts both pieces on the same clock. The January 2025 draft (docket FDA-2024-D-4488), jointly authored by CDRH, the biologics center (CBER), and the drugs center (CDER), covers total product lifecycle, or TPLC, risk management and marketing submission documentation for AI-enabled device software. The companion PCCP finalization tells sponsors what an acceptable change-control plan has to contain. Without both pieces, an AI developer can build a model, get it cleared, and then have no predictable way to update it.

RAPS, the regulatory affairs trade publication, reports the workplan covers 11 final guidance documents and three new drafts across the fiscal year, which begins 1 October 2026. The two A-list items are the load-bearing ones for AI medical-device makers; the rest of the list is largely standard CDRH housekeeping across diagnostics, radiology, and sterile device review.

The action clock for developers is tight. The public comment window on the agenda and the listed guidances closes 30 November 2026, with submissions routed through regulations.gov. For any company with a pending pre-submission meeting with FDA, a model in validation, or a quality management system already wired for retrospective change control, that eight-week window is the moment to align comments with the agency's PCCP framework and TPLC expectations. A pre-sub aligned to the 2025 draft is the cheapest way to find out where the agency will push back before the final rule lands.

The patient stake runs through the same mechanism. Adaptive AI in radiology triage, cardiac arrhythmia detection, and early-warning sepsis models is built to improve as it sees more cases. A finalized PCCP track means a model fielded at one site can be updated across a hospital network without each update restarting the clearance clock. The rules are still being written; the regulator is now telling sponsors when it expects to finish them, and giving them eight weeks to weigh in before the schedule hardens.

The watch item is whether the two A-list final guidances ship in the same fiscal quarter, or whether the PCCP finalization slips and leaves developers with a lifecycle rule but no usable change-control lane. The workplan sets the target; the docket and the FY 2028 agenda will show whether the agency hits it. STAT+'s Health Tech newsletter also re-reported the agenda this week, but the primary signal originates with CDRH's own published workplan.