First Edition — August 2026
AI is already inside your GMP processes. Govern it before an inspector asks.
The first operational guide with a toolkit on Artificial Intelligence in GMP environments: how to inventory, classify, validate and keep AI/ML components under control today — under the current Annex 11 — and how to prepare for the new Annex 22 without chasing the draft. Official sources verified on 17 August 2026.
13 chapters + 3 appendices6 editable toolsIT + EN3 case studiesLifetime updates
🧠
The complete framework
From the inventory of AI/ML components to drift monitoring in operation: classification decision tree, intended use, acceptance criteria, test data independency, explainability, suppliers.
🧰
The operational toolkit
3 Excel workbooks with formulas (the calculated decision tree, the gap assessment pre-populated on the draft, the metrics matrix with PASS/FAIL) + 3 Word templates with worked examples.
🔁
Lifetime updates
When the final Annex 22 is published, you automatically receive the updated revision by email. Buy once, stay aligned.
The toolkit: 6 ready-to-use tools, included in the price
Not empty shells: every file contains instructions, worked examples consistent with the guide's chapters, formulas and dashboards.
EXCEL
AI System Inventory & Applicability
Inventory your AI/ML components: the decision tree in formulas returns the classification outcome (A/B/C/D) for each system, with a dashboard.
EXCEL
Annex 22 Readiness & Gap Assessment
30 expectations pre-populated from the draft's 10 chapters, with gaps, risk, actions and a progress dashboard. A picture of your readiness in one day.
EXCEL
AI Risk & Acceptance Criteria Matrix
From the harm of a false negative to the right metric: thresholds, baseline, subgroups, formula-driven PASS/FAIL and a confusion matrix helper.
WORD
Intended Use & AI System Description
The founding document of every GMP AI system: input space, limits, subgroups, the role of the human — with a worked visual inspection example.
WORD
AI Test Plan Template
Dataset, representativeness, independency with an anti-leakage checklist, execution protocol, explainability and confidence, iteration management.
WORD
AI Supplier Assessment Questionnaire
46 questions in 8 sections for the vendor selling (or switching on) your AI: training data, versioning, metrics, use of your data, audit support.
You will use it tomorrow morning
| "Do we have AI in our systems?" — nobody can answer | AI System Inventory + the Chapter 4 decision tree |
| A new vision/ML system to validate | Intended Use + Risk & Acceptance Matrix + AI Test Plan (Chs. 5–8) |
| Your vendor announces an AI feature in the next upgrade | AI Supplier Questionnaire + Ch. 10 |
| The team wants an LLM assistant on SOPs | Outcome C classification + case study CS3 + human-in-the-loop |
| Understanding how far you are from the draft Annex 22 | Ch. 11 (6-step method) + Readiness Gap Assessment + roadmap |
Rigour, not hype: a draft is a draft
Annex 22 was released for consultation on 7 July 2025; the consultation closed in October 2025 and no final text has been published (the exclusions on dynamic, probabilistic and generative AI are under review after the EMA workshop of June–July 2026). This guide labels every statement — CURRENT, DRAFT, EXPECTED DIRECTION, BEST PRACTICE — and uses the draft for readiness only: today's compliance baseline remains Annex 11, and the guide shows you how to apply it to AI systems now.
Inside the guide
- Framework and principles: AI in GMP processes without the hype (vision, EM trending, predictive maintenance, LLMs), the exact regulatory status as of 17 August 2026, roles and RACI — including the data scientist as a role to govern
- The operational framework: classification decision tree (static/dynamic, deterministic/probabilistic, critical/non-critical), verifiable intended use, metrics and acceptance criteria with the confusion matrix explained on real numbers, test data independency and anti-leakage, explainability and confidence, drift monitoring and retraining as change control, purchased AI and suppliers
- Readiness in practice: Annex 22 readiness assessment in 6 steps with a worked example and roadmap, 3 full case studies (visual inspection, EM anomaly detection, LLM assistant), "What an Inspector May Ask — AI edition" across 8 areas
- Appendices: source register with URLs and verification date, 40-entry AI-GxP glossary, toolkit overview
| Format | Professional A4 PDF (71 pages IT / 73 pages EN) + toolkit as editable Excel/Word files |
| Edition | First — August 2026 (v1.0) · official sources verified on 17 August 2026 |
| Languages | Italian and English — pick your variant above (the toolkit ships in the variant language) |
| Updates | All future releases included for life, by email · dedicated revision upon publication of the final Annex 22 |
| Delivery | Instant digital download after payment |
Who it is for: QA Managers, CSV/CSA and Digital Quality Managers, IT Quality, Data Integrity Managers, System Owners, Process Owners, IT, data teams in regulated environments, Regulatory Affairs, QPs, auditors and GxP consultants.
GuideGxP is an independent publisher, not affiliated with the regulatory authorities cited. The guide does not replace official regulations: it systematically distinguishes between current requirements and draft proposals; content relating to documents not yet final — including the draft Annex 22 — is provided solely to support readiness assessment and must not be read as requirements already in force.
