Life Science & Pharma: GxP - AI

Draft EU GMP Annex 22, FDA's draft AI guidance for drugs and biologics, the joint FDA–EMA good AI practice principles, the EU AI Act, and related enforcement.

Documents
11
Authorities
European Commission, ICH, FDA, EMA, European Union, ISPE, MHRA, NMPA
Legal weight
Law & regulation, GMP guide & annexes, Guidance & Q&A, Industry good practice, Warning letters
Links on the map
3
Crosswalks
2
Next expected
Target 31 Dec 2026: Annex 22 · AI
How to read this page

Details: Guide Definitions: every term and label explained

Start with

The most connected documents in this area, so a good place to begin.

  • EMA AI reflection paper Final 2 links. EMA's current thinking on AI and machine learning for human and veterinary medicines, covering principles that apply at any stage of a medicine's lifecycle, from drug discovery to after authorisation.
  • HMA-EMA AI workplan Final 2 links. Joint HMA and EMA workplan for artificial intelligence in medicines regulation, 2023 to 2028.
  • EU AI Act Effective · phased 1 link. Horizontal EU law on AI systems, applying in phases.
  • ICH Q9(R1) Final 1 link. The international guideline on quality risk management.

Latest in this area

The five newest dated changes. All changes in this area · Timeline

  • 2 Aug 2026EU AI Act: Article 50 transparency duties apply (Art. 50(2) marking for systems already on the market: 2 Dec 2026, per the Omnibus)Effective · phased
  • 27 Jul 2026EU AI Act: Digital Omnibus on AI, Regulation (EU) 2026/1744, in force (published in the Official Journal 24 Jul 2026); defers high-risk obligationsEffective · phased
  • 30 Jun 2026Annex 22 · AI: EMA multistakeholder workshop on comments and generative AI (30 Jun–1 Jul)Draft
  • 2 Apr 2026FDA AI warning letter: Warning letter 320-26-58 issued (2 Apr 2026)Warning letter
  • 2 Apr 2026NMPA AI opinions: Issued by NMPAFinal

All 11 instruments

Weight means how binding the document is. Weight explained

Life Science & Pharma: GxP - AI: every instrument, with authority, legal weight, status, Category and last verified date
InstrumentAuthorityWeightStatusCategory · Sub CategoryLast verified
FDA AI guidance (drugs)FDAGuidance & Q&ADraftAI · GuidanceNot yet
EU AI ActEuropean UnionLaw & regulationEffective · phasedAI · RegulationNot yet
Annex 22 · AIEuropean CommissionGMP guide & annexesDraftAI · AnnexNot yet
FDA AI warning letterFDAWarning lettersWarning letterAI · EnforcementNot yet
FDA–EMA AI principlesFDA, with EMAGuidance & Q&AFinalAI · PrinciplesNot yet
ICH Q9(R1)ICHGuidance & Q&AFinalQuality foundations · GuidelineNot yet
MHRA AI regulation paperMHRAGuidance & Q&AFinalAI · GuidanceNot yet
NMPA AI opinionsNMPAGuidance & Q&AFinalAI · GuidanceNot yet
ISPE GAMP 5, 2nd ed.ISPEIndustry good practiceIndustry guideCSV · Industry guideNot yet
HMA-EMA AI workplanEMAGuidance & Q&AFinalAI · GuidanceNot yet
EMA AI reflection paperEMAGuidance & Q&AFinalAI · GuidelineNot yet

Map

Crosswalks in this area

A crosswalk matches clauses in one document with clauses on the same topic in another. Open one to see it row by row. "None" means that document has no matching clause, and "Difference" is our own short note, not a legal comparison.

EU GMP Annex 22 (draft) ↔ FDA AI guidance for drugs (draft) 11 topics

Annex 22 · AI compared with FDA AI guidance (drugs). Relationship types after NIST IR 8477 (STRM): equal, intersects, superset, subset, none. Drafted with AI from the official texts. Strength is a placeholder and is hidden until the SME Reviewer has reviewed the mapping. Both are drafts. Annex 22 covers GMP models. FDA covers AI that supports regulatory decisions, so their scopes differ.

Annex 22 · AITopicFDA AI guidance (drugs)RelationshipDifference
§1ScopeIIIntersectsAnnex 22 is limited to static, deterministic models in GMP and excludes generative AI. FDA covers models supporting regulatory decisions, and excludes discovery.
§3Intended use and context of useIV.A Steps 1–2IntersectsAnnex 22 requires a description and an SME. FDA frames a question of interest and context of use.
§2.3Risk basisIV.A Step 3IntersectsAnnex 22 uses patient safety, product quality and data integrity. FDA uses model influence and decision consequence.
§4Metrics and acceptance criteriaIV.A Step 4IntersectsAnnex 22 adds that criteria should be at least as high as the replaced process.
§5, §6Test data and independenceIV.A Step 4SupersetAnnex 22 is more detailed on independence of data and staff.
§7Test plan, execution and deviationsIV.A Steps 5–7IntersectsBoth require a plan, deviations recorded and a conclusion.
§8ExplainabilityNoneNo matchAnnex 22 has a section on feature attribution. FDA has no matching section.
§9Confidence scores and thresholdsNoneNo matchAnnex 22 has a section. FDA has no matching section.
§10.1–10.4Change and monitoring in operationIV.BIntersectsAnnex 22 lists controls. FDA discusses life cycle maintenance in certain contexts of use.
§3.3, §10.5Human reviewNoneNo matchAnnex 22 addresses human-in-the-loop use directly.
NoneEarly engagement with the regulatorIV.CNo matchFDA-specific section.
EU GMP Annex 22 (draft) ↔ FDA and EMA good AI practice principles 10 topics

Annex 22 · AI compared with FDA–EMA AI principles. Relationship types after NIST IR 8477 (STRM): equal, intersects, superset, subset, none. Drafted with AI from the official texts. Strength is a placeholder and is hidden until the SME Reviewer has reviewed the mapping. Annex 22 is a draft GMP text. The principles are high-level and non-binding.

Annex 22 · AITopicFDA–EMA AI principlesRelationshipDifference
§2.1People and expertisePrinciple 5IntersectsAnnex 22 names the roles. The principle is general.
§2.3Risk basisPrinciple 2IntersectsBoth apply effort according to risk.
§3Intended use and context of usePrinciple 4SupersetAnnex 22 is more detailed, including input sample space and subgroups.
§4, §7Performance assessmentPrinciple 8IntersectsThe principle covers the complete system, including human-AI interaction.
§5, §6Data and data documentationPrinciple 6IntersectsAnnex 22 focuses on test data. The principle covers data provenance and processing more widely.
§8ExplainabilityPrinciple 7IntersectsThe principle refers to interpretability and explainability in development.
§10Life cycle and monitoringPrinciple 9IntersectsBoth expect monitoring, including for drift.
§3.3, §10.5Human oversightPrinciple 1IntersectsAnnex 22 is about operator responsibility. The principle is about human-centric values.
NoneAdherence to standardsPrinciple 3No matchNo matching Annex 22 section.
NoneClear information for users and patientsPrinciple 10No matchNo matching Annex 22 section.
Relationship key
Equal
The two clauses cover the same ground.
Intersects
They overlap, but each covers something the other does not.
Superset
The first document's clause covers everything in the second, and more.
Subset
The first document's clause covers only part of the second.
No match
One document has no matching clause.

Relations on this map

The same links as the lines above, in words. What each link means