The Quality System Nobody Rebuilt Yet

The Quality System Nobody Rebuilt Yet

The generative AI docket gets the headlines, but the harder compliance problem in life sciences is already live. FDA’s Center for Devices and Radiological Health has authorized more than 1,000 AI-enabled devices, and the agency itself notes most are not generative AI at all marketscale.com. Those devices already operate under FDA’s Predetermined Change Control Plan framework, which lets manufacturers make pre-specified modifications without a full resubmission bipc.com. That sounds like relief. In practice it is a quality system obligation most manufacturers have not built for.

A PCCP is not a one-time filing. It is a living commitment to document anticipated modifications, the verification protocol for each one, and the data lineage that supports it, across the device’s total product life cycle meddeviceonline.com. Submissions now require ongoing generalizability testing to confirm that performance holds as the device encounters conditions beyond its original training and validation set, not just a design history file frozen at clearance. That is a fundamentally different artifact than the static DHF that ISO 13485 and 21 CFR 820 quality systems were architected to produce. Many organizations have an AI device playbook. Far fewer have a pipeline that generates continuous, versioned evidence rather than point-in-time proof marketscale.com.

Europe compounds the problem rather than simplifying it. MedTech firms are pushing to revise both the EU AI Act’s Digital Omnibus provisions and the MDR itself, arguing that overlapping obligations risk pushing innovation toward the US or Asia healthcare.digital. For US manufacturers building AI-enabled pipelines with European ambitions, that means designing evidence architecture against two moving targets at once: a US framework that expects continuous lifecycle documentation, and an EU framework still built around static conformity assessment even as it renegotiates its own AI-specific layer pharmaphorum.com.

The decision for regulatory and quality leaders is not whether to wait for FDA’s generative AI guidance to finalize. It is whether the QMS underneath every AI-enabled device already on the market, or in the pipeline, can produce the kind of continuous data lineage and performance monitoring that PCCPs already demand today. If that infrastructure does not exist, the gap shows up at the next planned modification, not at the next major guidance cycle.

Generative AI oversight will eventually get its framework. The 1,000-plus devices already operating under living change control plans do not have that luxury. Their evidence trail is due now, and most quality systems were not built to write it.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The argument is coherent and logically structured—moving from the PCCP framework’s requirements to the gap in existing QMS capabilities to the EU complication—but the central claim that ‘most quality
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but some sources are reused for multiple claims, which could be improved for clarity and specificity.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects FDA PCCP requirements, EU MDR/IVDR static conformity challenges, and the evolving EU AI Act landscape, though it does not explicitly cite ISO 42001.
Technical Accuracy	Llama	cleared. The article accurately describes the challenges of implementing a quality management system for AI-enabled medical devices under FDA’s Predetermined Change Control Plan framework and EU regulations.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies a core problem and supports it with external references, though the counterarguments are primarily implied rather than explicitly stated.
Novelty & Non-Duplication	Grok	held. Framing around unbuilt continuous-evidence QMS for existing PCCP devices is a usable synthesis angle, but the piece largely restates recent wire on FDA AI dockets, PCCPs, and EU overlap without clear
Validation	DeepSeek	cleared. The central claim that PCCPs impose a continuous quality system obligation not met by traditional static DHF-based systems is validated by cited industry sources describing the new requirements.

Sources cited: 11. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.