Does FDA regulate the model, or the software around it?

Both. In a medical device, the model is part of the device, its development, validation, and change management fall under the same design controls (21 CFR Part 820.30), IEC 62304 lifecycle, and ISO 14971 risk management as the rest of the software. Treating the model as separate from the device is the usual reason an AI submission struggles.

What is a Predetermined Change Control Plan (PCCP), and do we need one?

A PCCP lets you update a model after clearance, retraining, threshold changes, within bounds FDA has pre-authorized, without a new submission for each change. If your model will keep learning or be updated, a PCCP is how you keep it compliant without re-filing every iteration. We help scope and write it.

What is Good Machine Learning Practice (GMLP)?

GMLP is the set of guiding principles jointly issued by FDA, Health Canada, and the MHRA for the development of machine-learning medical devices, covering data quality, representativeness, training/test independence, and performance monitoring. We build to it so the model's development story holds up in review.

Can you bring a model our data scientists already built into a regulated product?

Yes. We bring an existing model under IEC 62304 and ISO 14971, establish the data and training provenance, help define the validation approach and acceptance criteria, and produce the verification evidence, so a research model becomes a regulated one.

Is our AI feature Software as a Medical Device?

It depends on intended use and the clinical decision it informs. We help you determine whether the feature is SaMD, a component of a device, or outside the definition, because that classification drives everything downstream.

Services · Teams adding AI/ML to a device, or building AI/ML SaMD

AI/ML in Medical Devices

RND builds and verifies AI/ML features in medical device software, Software as a Medical Device and AI-enabled instruments, under the approach FDA now expects: Good Machine Learning Practice, a Predetermined Change Control Plan for models that keep learning, and risk management that treats the model as part of the device from the first design review.

Talk to us about an AI/ML feature

SaMDFDA PCCPGMLPIEC 62304ISO 1497121 CFR Part 820.30510(k)De Novo

Model development & validationData pipelines under design controlAlgorithm change protocolMLOps for regulated softwareModel performance monitoring

Why does an AI/ML model need design controls?

A model that performs well in a notebook still has a regulatory gap to close before it becomes a cleared feature: data provenance, training and test independence, validation against a defined intended use, risk analysis of how the model can be wrong, and a plan for what happens when it is updated. RND has spent 25+ years turning software into regulated device software, AI/ML is the same discipline applied to the model.

What we do

Model development and validation, built and tested to Good Machine Learning Practice, with acceptance criteria tied to intended use.
Predetermined Change Control Plans, so a model can keep improving after clearance within FDA-authorized bounds.
Data pipelines under design control, provenance, versioning, and reproducibility that survive an audit.
Risk management for the model, ISO 14971 applied to model failure modes, drift, and generalization, feeding the device’s safety file.

Where the risk actually is

On a diagnostic system, an AI/ML failure usually shows up as the result, a wrong, missed, or misattributed output, not a dramatic malfunction. The threats are subtle: a training set that does not represent the deployed population, silent drift, a model confident where it should not be. We build the evaluation and monitoring that surface those before a reviewer, or a patient, does.

25+: 25+ years in business
15+ yrs: Engineers average 15+ years of experience

One partner, end to end

Software built alongside the instrument.

RND writes and verifies the software. Gener8 designs and manufactures the instrument, cartridges, and consumables around it, mechanical, electronics, optics, and microfluidics, under one roof. Software, device, and disposables from a single partner.

Explore Gener8 capabilities

Frequently asked questions

Does FDA regulate the model, or the software around it?: Both. In a medical device, the model is part of the device, its development, validation, and change management fall under the same design controls (21 CFR Part 820.30), IEC 62304 lifecycle, and ISO 14971 risk management as the rest of the software. Treating the model as separate from the device is the usual reason an AI submission struggles.
What is a Predetermined Change Control Plan (PCCP), and do we need one?: A PCCP lets you update a model after clearance, retraining, threshold changes, within bounds FDA has pre-authorized, without a new submission for each change. If your model will keep learning or be updated, a PCCP is how you keep it compliant without re-filing every iteration. We help scope and write it.
What is Good Machine Learning Practice (GMLP)?: GMLP is the set of guiding principles jointly issued by FDA, Health Canada, and the MHRA for the development of machine-learning medical devices, covering data quality, representativeness, training/test independence, and performance monitoring. We build to it so the model's development story holds up in review.
Can you bring a model our data scientists already built into a regulated product?: Yes. We bring an existing model under IEC 62304 and ISO 14971, establish the data and training provenance, help define the validation approach and acceptance criteria, and produce the verification evidence, so a research model becomes a regulated one.
Is our AI feature Software as a Medical Device?: It depends on intended use and the clinical decision it informs. We help you determine whether the feature is SaMD, a component of a device, or outside the definition, because that classification drives everything downstream.

Last reviewed October 1, 2026.