chapter 05 · projection

Future scope: 2026-2035.

The governance pen is moving. EU AI Act high-risk obligations land in two waves (Annex III · 2 Aug 2026; Annex I · 2 Aug 2027), AI-specific GMP guidance finalises the AI/ML manufacturing surface, ISO/IEC 42001 matures into the AI management-system anchor, and QMS converges with AIMS. Confidence is high where regulator text is drafted · medium where pilots run ahead of inspector acceptance · low where the policy direction is contested.

/ 01

The four forcing functions, 2026-2030.

EU AI Act · AI-specific GMP guidance · QMSR · ISO 42001

Four regulatory texts already in force or in late draft will reshape the governance surface through 2030. None of them is hypothetical. The dates below are statutory or in the published timetable.

2 August 2026

EU AI Act Annex III: high-risk applicability.

Regulation (EU) 2024/1689 entered into force 1 Aug 2024. Annex III high-risk obligations (medical devices on the AI Act overlay, safety components, biometrics, critical infrastructure) become enforceable for general-purpose deployers and providers. Conformity assessment, post-market monitoring, technical documentation, registration in the EU AI database.High

2 August 2027

EU AI Act Annex I: full applicability.

AI systems linked to harmonised product legislation, including medical devices and IVDs, should be assessed against the AI Act's phased applicability and conformity-assessment route. Notified-body capacity remains a practical watch item.High

Q4 2026 - Q2 2027

EU GMP AI-related GMP expectations watch.

EU GMP AI/ML expectations should be tracked through official EudraLex and EMA/EC publications; iFeed does not treat unpublished guidance as settled text. The practical watch areas are model lifecycle, training-data governance, change control, qualification status, and human accountability for release decisions.High

2 February 2026

QMSR maturation begins.

21 CFR Part 820 became the Quality Management System Regulation effective 2 Feb 2026 (FDA final rule, 31 Jan 2024). First inspection cycle drives 483 pattern shifts: ISO 13485:2016 cross-references, design history file scope clarifications, supplier-control re-interpretations. The inspection pattern should be monitored as post-effective-date experience accumulates.High

December 2023 published · 2026-2028 maturation

ISO/IEC 42001 as an AI management-system reference.

AI Management System standard published in 2023. Certification and supplier-use patterns are still maturing, so iFeed treats 42001 as a governance-evidence reference rather than a settled procurement requirement.High

2026-2030

QMS + AIMS convergence.

Two parallel management systems collapse into one operating model. ISO 9001 + ISO 13485 + ISO/IEC 42001 + ISO/IEC 27001 + ISO 14971 risk become an integrated stack with shared CAPA, management review, internal audit. Full convergence may emerge first in larger organisations with mature QMS and AI governance teams.Medium

/ 02

The 483 pattern shifts, 2026-2030.

What FDA inspectors find changes when QMSR settles

QMSR (eff. 2 Feb 2026) restated 21 CFR 820 with explicit cross-references to ISO 13485:2016. Inspection practice should be monitored as FDA and industry gain experience with the QMSR vocabulary.

FY26 - FY28

Design history file scope clarifications.

Old 820.30 DHF expectations interpreted against ISO 13485 §7.3. Combination-product DHF-to-PQS bridging is a plausible internal-audit topic; it should not be presented as an official future FDA category.High

FY27 - FY29

Supplier-control re-baselining.

820.50 vs ISO 13485 §7.4 supplier evaluation. Internal audits should check documented supplier risk classification by criticality. Vendor-quality questionnaires become more structured.High

FY27 - FY30

Risk-management file citations.

ISO 14971:2019 risk-management file expected as part of QMSR. Internal reviews should check post-production risk loops, missing benefit-risk re-evaluation triggers.High

FY28 - FY30

AI / ML in QMS process.

Where AI is embedded in CAPA triage, document control, complaint handling: lack of validation, lack of human-in-the-loop, training-data governance gaps. a practical watch item as adoption grows.Medium

FY26 - FY29

Data integrity remains a recurring watch area.

Data-integrity weaknesses remain a recurring inspection-readiness concern. MHRA GxP DI (2018, refreshed pattern) remains the global reference. Audit-trail review procedures and evidence remain important.High

FY28 onward

Notified-body spillover.

Cross-jurisdiction supplier evidence may become relevant in due diligence, but any regulator-to-regulator use should be confirmed from official policy or inspection experience.Medium

/ 03

The continuous validation paradigm.

Point-in-time validation is being challenged

Computer-system validation has lived under a lock-and-validate model since 1997 (21 CFR Part 11). AI/ML systems break the model: the model evolves, the data drifts, the population shifts. Three regulator instruments together challenge point-in-time validation models.

FDA · finalised Dec 2024

PCCP as the bridge.

Predetermined Change Control Plan (FDA PCCP guidance) lets sponsors describe planned model changes upfront, reducing supplemental approvals. The template for how a non-deterministic system becomes "validated under a plan" rather than relying only on a point-in-time validation story.High

EU AI Act Art 17

Quality management system obligation.

High-risk AI providers are expected to operate a documented QMS covering data governance, change management, monitoring, post-market surveillance. Life-sciences teams should watch how this is interpreted in sector-specific audits and supplier reviews.High

EMA Reflection Paper 2024 · refresh 2027

Continuous monitoring obligation.

EMA Reflection Paper on AI in the medicinal-product lifecycle (Sep 2024 final) introduces "monitor and update" as a regulatory verb. future updates may clarify specific KPI or monitoring expectations.Medium

2028-2030

QMS event cadence shifts.

AI-driven processes may justify tighter review cadence, clearer monitoring thresholds, and stronger management-review inputs; exact frequency should be justified by risk and procedure.Medium

2027

Model-update lifecycle codified.

Drift detection, retraining triggers, and revalidation scope should be defined in the change-control strategy for AI/ML systems.High

2030+

Continuous compliance dashboards.

Continuous-compliance dashboards may become a useful readiness model, but teams should treat live-inspection access as a controlled and jurisdiction-specific decision.Low

/ 04

Regulator AI literacy programs.

FDA AI Office · EMA AI WG · MHRA AI Airlock

The asymmetry between regulator capacity and sponsor AI deployment is the major 2026-2030 risk. Three regulators are building public learning, sandbox, and policy programmes that can inform governance literacy.

FDA · CDRH Digital Health Center of Excellence

FDA AI Office formalisation.

CDRH Digital Health Center of Excellence (DHCoE) consolidating 2024-2026 into a cross-Center AI policy hub. Cross-Center AI policy harmonisation should be monitored through FDA publications. Predetermined Change Control Plan (PCCP) is the single most influential deliverable so far.High

EMA · AI Working Group active since 2023

EMA AI WG · HMA-EMA Big Data Steering Group.

Reflection Paper on AI in the medicinal-product lifecycle (Sep 2024 final). 2025-2026 workplan covers data integrity, training-data governance, model-card requirements. Future updates should be monitored.High

MHRA · AI Airlock launched May 2024

MHRA AI Airlock.

Regulatory sandbox for AI medical devices. 5 candidate technologies in pilot wave 2024-2025. Findings publicly reported. Offers a useful sandbox model for other regulators to observe.High

PMDA · emerging

PMDA AI evaluation framework.

PMDA AI consultation pathway active since 2023. Any formal AI assessment guidance should be monitored through PMDA publications; lifecycle change control is the practical watch area.Medium

ICH · under Assembly review

ICH AI/ML reflection paper.

ICH Assembly initial discussion 2024. Cross-region harmonisation remains a watch area; iFeed should wait for official ICH topic adoption or concept papers before treating this as a programme.Medium

WHO · LMIC focus

WHO AI for health ethics · governance.

WHO ethics and governance guidance remains an important public-health reference. Its effect on ANVISA, CDSCO, and other agencies should be source-checked before being stated as a regulatory pathway.Medium

/ 05

The regulator governance maturity model.

An iFeed interpretation layer for 2028 readiness

A five-stage maturity model is a useful iFeed interpretation layer, not an official regulator model. It helps teams ask whether their QMS is reactive, managed, risk-driven, data-driven, or predictive. This mirrors what ICH Q10 introduced for pharmaceutical quality systems but extends it to the AI-augmented operating model.

Stage 1 · reactive
Compliance-driven.
Most pre-2025 organisations

QMS exists because the regulator requires it. Documents-on-paper culture. CAPA backlog grows. Inspector fatigue inevitable.

Stage 2 · managed
Process-driven.
2020-2026

Defined SOPs, measured deviations. Limited risk-based thinking. CAPA closed within timelines but effectiveness review is patchy.

Stage 3 · defined
Risk-driven QMS.
2024-2028

ICH Q9(R1) embedded. Risk register live, reviewed quarterly. CAPA effectiveness verified. A target state for mature teams.

Stage 4 · quantitatively managed
Data-driven QMS.
2027-2031

KPIs continuous. Trend analysis automated. Management review based on dashboards, not slides. ISO 42001 certifiable.

Stage 5 · optimising
Predictive QMS + AIMS.
2030+

Predictive quality. Drift detected before threshold breach. Model performance and process performance integrated. Live access, if used, should be controlled and jurisdiction-specific.

Inspector pressure
Stage 3 floor.
By 2028

iFeed interpretation: organisations that remain reactive or document-only will face higher readiness risk as inspections and supplier audits become more data-driven.

/ 06

The 2030+ governance landscape.

What inspectors will look for · what sponsors may need to produce

By 2030 the governance surface is structurally different from today. Five durable changes are visible in the regulator workplans now.

2026
QMSR baseline · ISO 42001 audit market opens.

Baseline: QMSR effective 2 Feb. ISO/IEC 42001 first independent certifications. EU AI Act Annex III applicability 2 Aug. The forcing-function year.

2027
EU AI Act Annex I · AI-specific GMP guidance endorsement.

2 Aug Annex I full applicability. AI-specific GMP and EMA AI updates should be monitored through official sources.

2028
483 pattern resettles.

QMSR-era inspection findings stabilise. AI-in-QMS becomes a watch item for internal audit, supplier control, and management review.

2029
QMS + AIMS integration visible.

Larger organisations operating one integrated stack. Quarterly management review with integrated KPIs. Smaller sponsors lag by 2-3 years.

2030
Continuous-compliance dashboards · ICH AI/ML M-series.

Continuous-compliance dashboards, ICH AI/ML harmonisation, and PMDA AI guidance remain watch items until official publications mature.

2032
Stage-3 QMS floor enforced.

Supplier due-diligence and internal readiness may increasingly ask for maturity self-assessment.

2035
Predictive QMS standard in major sponsors.

Highly mature organisations may move toward integrated QMS + AIMS architectures; iFeed should track this through public case studies, standards, and regulator material.

/ 07

Open questions through 2035.

Where the policy direction is contested

Three projection categories are not yet settled. Confidence is medium-to-low because the regulator pen is still moving, industry pilots are running ahead of inspector position, and statutory text leaves open interpretive room.

Generative AI in QMS

Authoring vs. review.

Generative AI in CAPA, deviation, and validation records should be treated cautiously. Human accountability, traceability, source control, and review evidence are the immediate governance questions; later regulator positions should be source-checked.Low

Foundation models · GPAI

GPAI obligations stack.

EU AI Act GPAI obligations apply to providers under the regulation; downstream deployer questions should be checked against official guidance and sector use case.Medium

FDA-EMA · mutual reliance

Mutual reliance on AI inspection.

Mutual reliance for AI-related inspection evidence is a watch area, not a current operating assumption.Low

LMIC pathway

WHO PQ AI conformity.

WHO, ANVISA, and CDSCO positions on AI-enabled submissions should be monitored through official guidance. Do not assume timing until published.Medium

ISO 42001 + ISO 9001 + ISO 13485

Single-audit model.

Whether one accredited audit body can issue a single combined certificate covering 9001 + 13485 + 42001. Industry pressure high. Combined-audit models should be checked with accreditation bodies and certification partners before being treated as available.Medium

Inspector capacity

The capacity bottleneck.

Regulator and notified-body capacity is a practical market watch item. Public claims should cite current workforce and notified-body sources before stating exact percentages or timelines.High

/ S

Source register.

official anchors · interpretation separated
FDA / QMSR

QMSR final rule.

Federal Register final rule amending the Quality System Regulation; effective date and FDA-specific overlays should be read from the rule text.

eCFR

21 CFR Part 820.

Current legal text for FDA device quality-system requirements; use this as the live clause anchor for QMSR references.

ICH

Q9(R1) quality risk management.

Step 4 guideline for quality-risk-management concepts, subjectivity, formality, and knowledge management.

ICH

Q10 pharmaceutical quality system.

Pharmaceutical quality-system reference for management responsibility, lifecycle quality, CAPA, and continual improvement.

ICH

Q12 lifecycle management.

Step 4 guideline for established conditions, post-approval change management, and product lifecycle management.

eCFR

21 CFR Part 11.

Electronic records and electronic signatures rule; use with predicate-rule context and FDA scope guidance.

FDA

Data integrity Q&A.

FDA questions-and-answers guidance on data integrity and CGMP; useful for ALCOA+ and audit-trail interpretation.

European Commission

EudraLex Volume 4.

Official EU GMP page for Annex 11 and related GMP annexes; AI-specific GMP claims should be checked here before publication.

EUR-Lex

EU AI Act.

Regulation (EU) 2024/1689 official text; used for AI Act timing, high-risk system references, and governance boundaries.

ISO

ISO/IEC 42001.

AI management-system standard landing page. Full standard text is paid; public iFeed content should not quote unavailable clauses.

FDA

Computer Software Assurance.

FDA guidance PDF for production and quality-system software; useful for CSA and CSV evidence-readiness discussion.

ISPE

GAMP 5, second edition.

Industry guidance landing page. Treat as implementation guidance, not a regulation; full guide access is controlled by ISPE.