What is the primary regulatory concern with adaptive algorithms?

The primary regulatory concern is ensuring safety and effectiveness are maintained as the algorithm evolves. This involves controlling the types of changes, monitoring performance, and managing potential biases or unintended behaviors introduced by new data.

How does the EU MDR address adaptive algorithms?

The EU MDR categorizes all software as medical devices, and adaptive algorithms fall under these regulations, requiring manufacturers to demonstrate conformity with Annex I General Safety and Performance Requirements. The focus is on robust quality management systems, risk management, and clinical evaluation throughout the device's lifecycle, including any post-market changes.

What role does real-world performance monitoring play for adaptive algorithms?

Real-world performance monitoring is crucial for adaptive algorithms to detect performance drift, identify new biases, and ensure the algorithm continues to meet its intended purpose safely and effectively over time. This data often informs retraining cycles and risk assessments.

All terms

Software & AIAI / ML in Devices Software Lifecycle

Locked vs. Adaptive Algorithm

Distinction between models whose behavior is fixed at release vs. those that continue learning.

Reviewed by Christian Espinosa, Founder, Blue Goat CyberLast reviewed September 19, 2026

Definition

A locked algorithm produces the same output for the same input every time after release. An adaptive algorithm changes its behavior over time as it is exposed to new data, either continuously or in scheduled retraining cycles. Most authorized AI devices are locked: they may be updated, but each update is a planned, validated release. Truly adaptive, continuously learning devices raise questions about how to verify performance, so regulators expect strong controls. In the US, a PCCP lets manufacturers pre-specify the changes an algorithm can undergo, and the EU AI Act requires substantial changes to high-risk systems to go through a new conformity assessment unless predetermined.

What the regulation says

The FDA's guidance, 'Good Machine Learning Practice for Medical Device Development' and the 'Predetermined Change Control Plan' (PCCP) framework, differentiates between locked and adaptive algorithms. For locked algorithms, regulatory review focuses on the algorithm's performance at the time of submission. For adaptive algorithms, regulators assess the manufacturer's control over the modification protocol, the 'Algorithm Change Protocol,' to ensure predictable and safe evolution.

What this means in practice

FDA's PCCP framework lets manufacturers prospectively define and validate the boundaries of allowable adaptive changes so retraining does not require a new submission for each model update.

Examples

A diagnostic AI that learns to identify new patterns in medical images as it processes more patient scans is an adaptive algorithm requiring a PCCP.
An insulin pump algorithm that adjusts dosing recommendations based on a patient's continuously acquired glucose data and activity levels is an adaptive algorithm.
A risk stratification tool that uses a fixed set of patient demographics and historical health records to predict disease likelihood is a locked algorithm.

Common pitfalls

•Failing to establish clear performance bounds and risk mitigations for adaptive algorithm changes can lead to regulatory non-compliance.
•Treating an adaptive algorithm as a locked algorithm for validation purposes will result in insufficient oversight of its evolving behavior.
•Not documenting the data used for training and retraining, along with corresponding version control, can create significant auditability issues.
•Assuming all adaptive algorithm changes are minor and do not impact safety or effectiveness can lead to patient harm and regulatory scrutiny.

Frequently asked questions

Yes, an adaptive algorithm might be 'locked' at a specific version after extensive validation, especially if its environment stabilizes or further adaptation introduces undue risk. This transition typically requires clear documentation and justification.

Cross-references

Uses

Concepts or artefacts this term builds on.

Predetermined Change Control Plan(PCCP)

Grouped by theme

Editor's picks

· Hand-selected related concepts

Software & AI

AI/ML-Enabled Medical Device

Medical device that uses artificial intelligence or machine learning to perform its intended use.

Software & AI

Good Machine Learning Practice(GMLP)

Guiding principles for the development of AI/ML-enabled medical devices.

Software & AI

Predetermined Change Control Plan(PCCP)

FDA mechanism to pre-authorize specific modifications to AI/ML-enabled devices.

More in Software & AI

· Same category

Software & AI

AAMI TIR45(TIR45)

AAMI Technical Information Report providing guidance on applying Agile software development practices within an IEC 62304-compliant medical device software lifecycle.

Software & AI

Adversarial Robustness

Resilience of an ML model to inputs deliberately crafted to cause misclassification.

Software & AI

Algorithm Change Protocol(ACP)

The detailed procedural section of a PCCP that specifies how planned modifications to an AI/ML model will be developed, validated, and implemented.

Software & AI

Algorithmic Bias and Fairness

Systematic differences in model performance across demographic or clinical subgroups.

Cited by

Where this term appears across MedTech Terms.

Ecosystems (2)

Latest in MedTech

Sources

3 sources

Every citation below opens the original document. Each is graded against our source-tier hierarchy so you can see what rests on binding law versus commentary.

Tier 2Regulator guidance and consensus· 3

Link health: 3 verified· last checked 2026-06-20

MDCG·1FDA·2

Inline markers like [1] jump to the matching reference above.

My progress

All terms0 / 513

Software & AI0 / 31

Saved on this device, and to your account when signed in. View studied terms

Full glossary

Browse all terms A to Z, each with a short definition.

On this term

Category: Software & AI
Sources: 3
Updated: 9/19/2026

Compare with another term

Learn in 60 seconds

Distinction between models whose behavior is fixed at release vs. those that continue learning.

·FDA's PCCP framework lets manufacturers prospectively define and validate the boundaries of allowable adaptive changes so retraining does not require a new submission for each model update.
·An adaptive algorithm changes its behavior over time as it is exposed to new data, either continuously or in scheduled retraining cycles.
·Most authorized AI devices are locked: they may be updated, but each update is a planned, validated release.

Remember this

Watch out: Failing to establish clear performance bounds and risk mitigations for adaptive algorithm changes can lead to regulatory non-compliance.

Related terms

You may also need

Auto-suggested from Software & AI and shared keywords.

All Software & AI terms

From the Blue Goat network

Related resources and services on this topic.