The ICH Working Group on AI: Three Real Documents Pharma Quality Should Read This Month

The ICH Context for AI

The International Council for Harmonisation has been the central mechanism for converging pharmaceutical regulatory expectations across FDA, EMA, PMDA, Health Canada, MHRA, and other major regulators since the 1990s. ICH guidelines that reach Step 5 — the final implementation step — become essentially binding across these jurisdictions, providing the operational foundation for pharma sponsors to work from a single set of expectations rather than parallel jurisdictional requirements.

ICH’s posture on AI specifically has, until recently, been worked through adjacent guidelines rather than through a dedicated AI guideline. The reasoning is structural: AI is a technique used across many pharma activities, not a discrete activity in itself. ICH has therefore extended existing guidelines to accommodate AI — most prominently through M15 on model-informed drug development — rather than starting a new guideline specifically for AI.

The practical consequence is that pharma quality leaders looking for an “ICH on AI” document in the same form as ICH Q9 (Quality Risk Management) or ICH Q10 (Pharmaceutical Quality System) will not find one. What they will find is a coordinated set of documents that, taken together, articulate the ICH-track convergence on AI. Reading these together is the right discipline; treating each in isolation misses the harmonization picture.

This article focuses on the three documents that, as of May 2026, most concretely articulate the convergence picture: ICH M15, the FDA’s January 2025 credibility framework draft guidance, and the FDA/EMA joint Good AI Practice principles. Other documents are coming — including the EMA’s reflection paper on AI in the medicinal product lifecycle and continued ISPE engagement — but these three are the most immediately operational for pharma quality leaders this month.

Document 1: ICH M15 on Model-Informed Drug Development

ICH M15, formally titled “General Principles for Model-Informed Drug Development,” is the most consequential ICH-track development for AI in pharma to date. It was endorsed for public comment by the ICH Assembly in November 2024, reached Step 2b for European consultation, and reached formal adoption in January 2026. The guideline is accessible through the EMA’s scientific guideline page for ICH M15 and through corresponding regulator pages.

The guideline matters for AI for a specific reason. Its definition of Model-Informed Drug Development (MIDD) explicitly includes “computational modeling and simulation methods (e.g., agent-based models, artificial intelligence/machine learning) that integrate nonclinical and clinical data, prior information, and knowledge.” This is the first time AI/ML has been explicitly recognized within an ICH guideline as a category of regulated drug development activity. The recognition is operationally significant: it brings AI/ML used in drug development under the harmonized ICH structure rather than leaving it to jurisdictional fragmentation.

What M15 actually covers:

• A harmonized assessment framework for MIDD evidence, including an assessment table for communication between sponsors and regulators
• Principles for the planning, evaluation, communication, and regulatory use of model-based evidence
• Categories of MIDD applications that the guideline explicitly addresses, including dose selection, exposure-response evaluation, pediatric extrapolation, PBPK-based DDI waivers, and virtual bioequivalence
• Terminology that aligns across the FDA, EMA, PMDA, and other ICH member regulators

As Certara’s FAQ on ICH M15 explains, the guideline does not specify AI-specific validation requirements; it instead establishes the harmonized framework within which AI use in MIDD is evaluated. This is consistent with ICH’s broader approach: the guideline articulates the framework, and jurisdictional regulators issue implementation guidance.

For pharma quality leaders, M15 has three immediate operational implications. First, documentation for AI/ML use in MIDD should align with the M15 vocabulary and structure, which produces submissions that satisfy multiple regulators from a common base. Second, quality teams should engage with the assessment framework explicitly, using M15’s structure as the documentation backbone for AI/ML model-based evidence. Third, the M15 adoption signals where ICH is heading more broadly: harmonized AI frameworks built on existing pharma quality scaffolding rather than parallel AI-specific structures.

Document 2: FDA’s Credibility Framework as ICH Bridge

The FDA’s January 6, 2025 draft guidance, Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products, accessible through the FDA guidance landing page, is the second document quality leaders should be reading. The connection to ICH is structural: the credibility framework is explicitly designed to align with international approaches and to bridge between domestic FDA expectations and the ICH harmonization track.

The framework’s seven-step structure — defining the question of interest, defining the context of use, assessing model risk, planning credibility activities, executing the plan, documenting evidence, and determining adequacy — is being designed to interoperate with the ICH M15 assessment framework. Quality leaders working from the credibility framework for FDA submissions and from M15 for international submissions will find that the structures fit together rather than conflicting.

The practical reading of the FDA document for ICH-aware quality leaders:

• Pay particular attention to the model risk assessment step, which is the structural counterpart to M15’s assessment table
• Note the scope clauses excluding drug discovery and operational uses from the credibility framework, which are FDA-specific scope clauses that may not perfectly align with ICH expectations as those develop
• Treat the framework as the FDA’s expression of an emerging international consensus, not as a US-only requirement that will need to be adapted for ex-US submissions

The FDA’s January 2025 press announcement for the credibility framework explicitly notes that the framework was developed with input from international counterparts, signaling that the document is intended to interoperate with the ICH harmonization track rather than to diverge from it.

Document 3: The FDA/EMA Joint Guiding Principles

The third document is the “Guiding Principles of Good AI Practice in Drug Development,” jointly developed by the FDA and EMA and accessible through the FDA Guiding Principles page. The ten principles articulated in the document are:

1. Human-centric by design
2. Risk-based approach
3. Adherence to standards
4. Clear context of use
5. Multidisciplinary expertise
6. Data governance and documentation
7. Model design and development practices
8. Risk-based performance assessment
9. Life cycle management
10. Clear, essential information

The joint development of these principles by FDA and EMA is the most explicit signal yet of the harmonization track. The document is not formally an ICH document, but the principles align tightly with the structural logic of ICH guidelines: principles articulated at a level of generality that allows operational implementation while supporting cross-jurisdictional consistency.

For pharma quality leaders, the practical use of the principles is as a structural reference for QMS documentation. Each principle should be addressable in QMS extensions for AI: how does the framework ensure human-centric design, how does it implement risk-based approach, how does it adhere to standards, and so on. Documentation organized around the ten principles satisfies both FDA and EMA expectations from a common base and ages well as the harmonization track produces additional documents.

The principles also provide a vocabulary for cross-functional conversation. The “human-centric by design” principle, the “risk-based approach” principle, and the “lifecycle management” principle each translate into operational discussions that QA, IT, regulatory, and use case teams can have on a shared foundation. This vocabulary work is undervalued but materially useful in operationalizing the framework.

The Convergence Picture They Together Form

Read as a connected set, the three documents articulate a recognizable convergence picture. The key elements:

The convergence is real enough that pharma quality leaders can reasonably build a single AI governance framework that satisfies the requirements articulated across all three documents. The framework would be organized around the ten FDA/EMA principles, implement the credibility framework’s seven steps for context-of-use evaluation, and align documentation with the M15 assessment structure for MIDD applications.

Operational Implications for Pharma Quality

The convergence picture has several immediate operational implications.

Single framework, multiple jurisdictions. Pharma quality leaders no longer need to plan for materially divergent jurisdictional requirements on AI in drug development. The framework that satisfies the FDA credibility approach, the ICH M15 structure, and the FDA/EMA principles will satisfy the majority of regulator expectations for AI in MIDD and adjacent use cases. This is a significant simplification compared to the early-2024 landscape.

Documentation alignment matters more than documentation completeness. Quality teams that produce comprehensive AI documentation without aligning to the harmonization vocabulary will find their documentation more burdensome to navigate for inspectors and reviewers than teams that align with the harmonization structures. The alignment work is a small additional investment that pays back significantly.

Tier classification SOPs should reflect the harmonized risk language. “Model risk,” “context of use,” and “risk-based approach” are the vocabulary the regulators are converging on. Quality teams using divergent vocabulary in their tier classification work will need to do translation work for each inspection or submission. Aligning the vocabulary now reduces ongoing friction.

Cross-functional governance committees should be reading the three documents together. The convergence is most useful when the cross-functional steering committee for AI governance has internalized it as a coherent picture. Reading them as separate items in different committees produces fragmentation; reading them together produces the right operational alignment.

Inspection readiness materials should reference the harmonization track. Inspectors and reviewers who themselves are working from the harmonized vocabulary will find documentation that references the FDA/EMA principles, the credibility framework, and M15 more readily navigable than documentation that uses bespoke organizational vocabulary. Inspection-readiness summaries should anchor in the harmonization language.

What’s Coming Next From ICH

The ICH track will continue to produce material output over the next twelve months. Several developments to watch.

First, regional implementation guidance for ICH M15 from FDA, EMA, PMDA, and Health Canada. Each regulator will issue implementation guidance specifying how M15 will be applied within their jurisdiction. These regional implementations will resolve some of the ambiguities visible in the harmonized document.

Second, finalization of the FDA’s January 2025 credibility framework guidance, expected during 2026. The finalized guidance will likely incorporate consultation feedback and may further align with the ICH M15 vocabulary.

Third, EMA’s Annex 22 finalization, which while EU-specific will be coordinated with the ICH harmonization track and will affect multinational manufacturers significantly.

Fourth, ongoing CDER work through the FRAME initiative, which CDER described in its March 2026 ICH proposal as a vehicle for new ICH guideline work to facilitate adoption of advanced pharmaceutical manufacturing including AI. This signals additional ICH-track work specifically on manufacturing, complementary to M15’s drug development focus.

Fifth, continued ISPE GAMP work, including the standalone GAMP Guide on Artificial Intelligence published in July 2025, which provides industry-side operational guidance complementary to the regulatory documents. As BioProcess Online’s coverage of the ISPE framework indicates, this guide extends GAMP 5 principles specifically to AI-enabled computerized systems across the lifecycle.

Pharma quality leaders monitoring these developments through normal channels — official regulator pages, ICH publications, ISPE engagement, industry working groups — will see the harmonization track continue to mature. The work being done now to align QMS extensions with the converging framework will age well into this evolving environment, while parallel jurisdictional approaches will face mounting alignment debt. The strategic posture is clear: build for the convergence, not for the fragmented pre-2025 picture.

What the harmonization is not yet addressing

Reading the three documents together also clarifies what the harmonization is not yet addressing. Three gaps are worth flagging because they will shape the next phase of work.

First, the harmonization does not yet address generative AI specifically. ICH M15’s MIDD scope, the FDA credibility framework’s drug development focus, and the FDA/EMA principles’ broad framing all leave generative AI use cases — drafting, retrieval, summarization, content generation — relatively under-articulated. The EMA Annex 22 addresses generative AI by excluding it from critical manufacturing functions, but the broader question of how generative AI should be governed across drug development is not yet harmonized. This is likely to be the next major area of harmonization work, and quality leaders should expect material publications on generative AI specifically over the next twelve to eighteen months.

Second, the harmonization does not yet articulate detailed expectations for ongoing performance monitoring. The principles and credibility framework both reference lifecycle management, but the operational specifics of what monitoring is required, at what cadence, and what triggers regulatory action are not yet articulated in harmonized form. Quality leaders should expect this gap to close, but the work is more likely to come through ISPE GAMP and PDA technical reports than through additional ICH documents in the near term.

Third, the harmonization does not yet address the auditor and inspector competency model for AI. The documents are designed for use by qualified inspectors, but the question of how inspectors themselves develop and maintain AI competency is left to jurisdictional regulator workforce development. This gap creates uncertainty for sponsors about the practical sophistication of AI-related inspection findings during the early implementation period.

How to operationalize the convergence in your QMS extensions this quarter

For pharma quality leaders looking for specific actions to take in the coming quarter, the convergence picture suggests a concrete sequence. Start by mapping your existing AI use case inventory against the FDA/EMA ten principles and identifying which principles are well-addressed by your current QMS and which are underdeveloped. Then for each AI use case in scope of regulatory submissions, draft the credibility framework’s seven-step analysis using the language from the FDA draft guidance. Finally, for use cases that fall within ICH M15’s MIDD scope, align documentation with the M15 assessment table structure so that the same evidence supports submissions to multiple regulators.

This sequence — principles alignment, then credibility framework application, then M15 structural alignment — produces QMS extensions that are coherent across the three documents rather than fragmented. The work is real but bounded, and a focused quarter of dedicated effort can produce material progress. Quality leaders who report this progress to their cross-functional steering committee establish a foundation for continued work as additional harmonization documents emerge.

Working with the regulators directly

A final operational point worth emphasizing: the harmonization track is ongoing, and the regulators are actively soliciting input. The FDA’s external engagements page, the EMA’s stakeholder consultations, and the ICH public comment processes are mechanisms through which sponsor input shapes the next phase of harmonization. Pharma quality leaders who engage in these mechanisms — whether directly or through trade associations — both inform the framework’s evolution and gain early visibility into its direction. The engagement is not a one-way obligation; it is a two-way mechanism that both shapes the harmonization and accelerates the engaging organization’s own learning. Quality leaders who treat these engagements as a strategic priority rather than a compliance overhead consistently develop more sophisticated frameworks than peers who engage minimally.

References & Sources