We are currently upgrading our website. Please bear with us if you are experiencing any loading issues.
APACMed
The Voice of MedTech in Asia
JOINT INTELLIGENCE PIECE
When Regulators Use AI:
What It Means for MedTech Teams
Artificial intelligence (AI) has transitioned from a futuristic discussion point to an operational reality within global regulatory authorities. While the MedTech industry’s focus has been on how to develop and regulate AI-enabled medical devices, a parallel revolution is occurring on the other side of the desk. Faced with an overwhelming volume of clinical protocols, safety reports, and expanding dossiers, regulators are implementing AI to manage workloads and reduce administrative bottlenecks.
These are two distinct developments, each with different implications for regulatory affairs professionals. Understanding them separately is essential before deciding what to do about either
This article focuses primarily on the first: how regulators are deploying AI internally to manage submissions, screen safety data, and accelerate decisions, and what that means for MedTech teams preparing dossiers. Where relevant, it also flags where the two threads intersect, particularly for manufacturers of AI-enabled devices who now face AI-enabled review.
A Snapshot of Jurisdictional Internal
Regulator AI Adoption
Artificial intelligence (AI) has transitioned from a futuristic discussion point to an operational reality within global regulatory authorities. While the MedTech industry’s focus has been on how to develop and regulate AI-enabled medical devices, a parallel revolution is occurring on the other side of the desk. Faced with an overwhelming volume of clinical protocols, safety reports, and expanding dossiers, regulators are implementing AI to manage workloads and reduce administrative bottlenecks.
These are two distinct developments, each with different implications for regulatory affairs professionals. Understanding them separately is essential before deciding what to do about either
This article focuses primarily on the first: how regulators are deploying AI internally to manage submissions, screen safety data, and accelerate decisions, and what that means for MedTech teams preparing dossiers. Where relevant, it also flags where the two threads intersect, particularly for manufacturers of AI-enabled devices who now face AI-enabled review.
What this means operationally:
AI is now screening submissions before a human reviewer ever opens the dossier. Structural gaps, missing metadata, and inconsistent module linkages are flagged immediately. A narrative that a human reviewer might follow despite ambiguity will not survive automated triage.
The APAC Landscape:
Uneven but Accelerating
Across Asia-Pacific, internal Al adoption in regulatory operations is real and growing, but maturity varies significantly by authority.
Singapore
(HSA)
Singapore’s HSA launched its AI-Enabled Medical Device Risk Classification Tool in March 2026, allowing sponsors to input a device name and intended use and receive an estimated risk classification before formal submission, a meaningful shift toward logic-driven pre-screening.
JAPAN
(PMDA)
Japan’s PMDA published its Action Plan for AI Use in Operations in October 2025, and as of 15 April 2026 confirmed the live deployment of its AI digital assistant, with officers using it for document creation, translation, and regulatory data retrieval. What was a plan six months ago is operational today.
CHINA
(NMPA)
China’s NMPA confirmed its first live internal AI deployment in January 2026: an LLM-based query tool enabling regulatory staff to interrogate large volumes of data through conversational dialogue.
Where The Two Threads Intersect:
Lifecycle Frameworks For AI-Enabled Devices
Regulators are not only using AI, they are simultaneously adapting the frameworks that govern AI-enabled devices themselves. Traditional frameworks assumed devices were “fixed.” However, machine-learning models introduce plasticity: performance that may evolve post-approval, and a degree of unpredictability. Regulators are responding by introducing mechanisms for planned post-market change without requiring full re-submission:
With the FDA's Predetermined Change Control Plan (PCCP);
With the EU's December 2025 MDR/IVDR amendment introducing PCCPs mirroring the FDA approach, alongside regulatory sandboxes and clarified AI Act/MDR/IVDR overlap guidance;
With Japan's IDATEN system similarly allowing post-market implementation of prespecified modifications through a simplified notification.
The first approval is no longer the finish line; for AI-enabled devices, it is the start of a managed lifecycle that regulators now expect to see planned for from day one.
Beneath the Surface:
Technical and Operational Risks
The integration of AI into regulatory workflows introduces a new category of operational risk: algorithmic volatility. A high-profile example is the FDA’s migration of its Elsa assistant from one underlying model to another. From a regulatory science perspective, this is not a routine software update. Changing the underlying model in a RAG system requires revalidating how the AI retrieves, interprets, and summarizes documents.
Re-validation and
Consistency
If a submission is reviewed during such a transition, the administrative record may contain inconsistent AI-generated summaries.
The Necessity of
Human Oversight
Even advanced models face internal criticism for “hallucinations”, generating fake citations or inconsistent outputs. This is why human oversight remains essential; in rare cases, a query may stem from an AI retrieval error rather than a gap in the dossier itself.
Emerging Concern:
Algorithmic Bias
At the level of regulatory AI tools, the concern of algorithm bias is more theoretical but worth monitoring: any system trained predominantly on submissions from large organisations in mature markets may over time develop structural preferences for those formats. No regulator has confirmed this risk in their own tools, but it is a question the industry should be asking.
What Teams Should Do Now:
Concrete Actions
The shift to AI-enabled regulatory review has concrete operational implications.
Regulatory professionals must adopt practical data-governance safeguards:
1
Audit your active submissions against machine-readability standards.
You don't need to know the exact algorithm used by your regulator to prepare well: what you need to know is that AI screening systems check for structural consistency, mismatches between modules, claims that do not trace to evidence, and missing metadata.
A human reviewer can follow a narrative with a gap; an AI tool flags it immediately. Ask whether your data structure is consistent across modules, claims are explicitly mapped to evidence, and metadata is complete for each target market. The gaps an AI screening system would flag become formal deficiencies. Find them before they become formal deficiencies.
2
Log and audit all regulatory interactions.
As AI-supported review systems become more integrated into regulatory operations, contemporaneous record-keeping becomes increasingly important. Maintain clear records of submissions, queries, clarifications, and responses throughout the review lifecycle. Structured documentation supports alignment, traceability, and efficient follow-up discussions, particularly in fast-moving or digitally enabled review environments. Structured documentation also helps teams maintain consistency across multi-market submissions and evolving digital review processes.
3
For manufacturers of AI/ML based SaMD
Map your compliance deadline exposure by market. Several AI-related milestones are active or imminent: the EU AI Act high-risk obligations apply from 2 August 2026, and the IMDRF draft Technical Framework for Artificial Intelligence Life Cycle Management is open for public consultation until July 2026. Teams with multi-market strategies should be engaging with the consultation now, not waiting for individual jurisdictions to transpose it. Understanding exactly where your portfolio stands against each deadline, in each market you operate in, is key.
Beyond these immediate steps, MedTech teams developing AI-enabled devices should engage now with the pilots, sandboxes, and pre-submission programs where regulatory expectations are being calibrated. The MHRA AI Airlock is live; EU regulatory sandboxes open in August 2026; the FDA TEMPO programme is open today. The teams that engage now will help shape the rules. Those that wait will adapt to rules they had no part in forming.
Beyond these immediate steps, MedTech teams developing AI-enabled devices should engage now with the pilots, sandboxes, and pre-submission programs where regulatory expectations are being calibrated. The MHRA AI Airlock is live; EU regulatory sandboxes open in August 2026; the FDA TEMPO programme is open today. The teams that engage now will help shape the rules. Those that wait will adapt to rules they had no part in forming.
Conclusion:
Thriving in the AI Era
The 2025–2026 period has made one thing clear:
AI is no longer just something regulators are asked to oversee in products. It is embedded in how they work.
For MedTech regulatory professionals, it means structured, machine-readable, traceable evidence, and defensible internal AI governance, aligned to AI-enabled regulatory expectations, across regions, on tighter timelines.
Readiness is no longer about awareness. It is about whether your regulatory strategy can meet the expectations of an AI-enabled regulator, while simultaneously demonstrating that your own AI-enabled product is governed to the standards those same regulators are now enforcing.
Author Bio
Patricia Teysseyre
Senior Regulatory Affairs Professional
AI is no longer just something regulators are asked to oversee in products. It is embedded in how they work.
For MedTech regulatory professionals, it means structured, machine-readable, traceable evidence, and defensible internal AI governance, aligned to AI-enabled regulatory expectations, across regions, on tighter timelines.
Readiness is no longer about awareness. It is about whether your regulatory strategy can meet the expectations of an AI-enabled regulator, while simultaneously demonstrating that your own AI-enabled product is governed to the standards those same regulators are now enforcing.