What the guide is
ISPE lists the GAMP Guide: Artificial Intelligence as a 290-page guidance document published in July 2025. ISPE describes it as a stand-alone guide designed to be used in parallel with GAMP 5: A Risk-Based Approach to Compliant GxP Computerized Systems (Second Edition, July 2022) and the other GAMP guides. It was sponsored by the GAMP Community of Practice Software Automation and AI Special Interest Group, and its leads describe its aim as setting out the overall 'what' layer for developing and using AI-enabled computerized systems. They write that the group had been flooded with questions asking how, if at all, AI could be validated.
The guide is a paid ISPE publication. This page summarises only what ISPE has published openly about it: the product page, the guide's public table of contents, an August 2025 iSpeak blog post by the guide's leads, and Pharmaceutical Engineering articles. It does not reproduce the guide's text, and anyone applying it should work from the purchased guide itself.
How the guide frames validation
ISPE frames the goal in terms of fitness for intended use. Its product page says the framework in this guide helps ensure that AI-enabled computerized systems are high-quality, effective, fit for their intended use and compliant with applicable regulations, and that it offers a robust, cost-effective approach when applied with expertise and good judgement.
The public table of contents places validation inside the guide's life cycle approach (Chapter 3), in a section titled AI-Enabled Computerized System Validation Framework and Critical Thinking, next to a section on specification and verification. Testing of AI-enabled computerized systems appears within the project phase appendix, and the guide has a separate appendix on critical thinking with considerations for each phase, supplier selection and inspection readiness. The content of the validation framework itself is in the paid guide; ISPE's public material names it but does not set it out.
Scope: which systems and which organisations
According to ISPE's product page, the approach is suitable for all types of computerized systems using AI, and applies to standard and configurable products as well as custom applications. Its stated aim is to safeguard patient safety, product quality and data integrity in GxP areas. The table of contents also lists sections on the EU AI Act, supplier aspects, software and hardware categories, model categories and AI maturity, including AI sub-system autonomy and adaptiveness. It also has an appendix on statutes, regulations, standards and guidance and a section on governance for achieving compliance; the public material lists these headings but not their content.
ISPE names regulated companies, suppliers, service providers and regulators as the audience, and the guide defines activities and responsibilities for both internal and external suppliers. A Pharmaceutical Engineering article by the guide's leads names GxP-regulated companies in pharmaceuticals, biologics, advanced therapy medicinal products or medical devices as examples of its stakeholders, says it is also relevant to early research outside traditional GxP-regulated areas, and says regulators and healthcare authorities may find it a valuable reference for understanding industry perspectives on best practice for AI. For AI in and as medical devices, the leads say the guide integrates device practices such as design reviews and device-specific risk management. The guide does not replace the device regulations that apply.
The risk-based approach
ISPE's public descriptions put quality risk management at the centre. The guide sets out AI-specific risk considerations and a framework for identifying risks and establishing suitable control strategies. It promotes a risk-based approach and the use of critical thinking, and encourages scalable life cycle activities. The table of contents shows a chapter on quality risk management and a science-based QRM appendix that covers, among other things, scaling the process, risk management through the system life cycle, hazard identification, selection of controls, residual risk and risk considerations when applying explainable AI. ISPE lists focusing effort by risk, choosing system designs that fit the organisation's experience, and better collaboration between regulated companies and suppliers among the benefits.
The life cycle phases in the table of contents
ISPE says the guide covers the AI-enabled computerized system life cycle from concept to retirement, with key supporting processes. Its public table of contents has one appendix per phase. The table below lists a selection of each appendix's section headings. ISPE's public material states no acceptance-criteria values, performance thresholds or pass/fail limits, and none should be inferred from this table.
| Phase (appendix) | Selected section headings in ISPE's public table of contents |
|---|---|
| Concept (P1) | Proof-of-concept structure; context of use and scope definition; drafting requirements; initial risk assessment and trustworthy AI considerations; feasibility assessment; proof-of-concept evaluation criteria |
| Project (P2) | Model requirement specification and model design space; case data set creation and data split; iterative experimentation; model release candidate selection; testing of AI-enabled computerized systems; reporting and release |
| Operation (P3) | Handover; system monitoring; incident and problem management; CAPA; operational change and configuration management; periodic review; security management |
| Retirement (P4) | Retirement planning, execution and reporting; lessons learned and improvements |
Data, models and people
Data gets particular attention. The leads write that the guide introduces fit-for-purpose data and data quality considerations, with an extended view of data governance, and note that data collected for its original purpose may not be fit for AI development. The table of contents has appendices on fit-for-purpose data and data quality, on data and model governance and management, and on IT infrastructure, including model monitoring and drift detection. The Chair's message in the January/February 2026 issue of Pharmaceutical Engineering says the guide addresses training data provenance, model acceptance criteria, performance monitoring and drift detection; the public material does not state what those criteria are.
The leads say the guide expands on the concept of dynamic systems from Appendix D11 of GAMP 5 (Second Edition), titled Artificial Intelligence and Machine Learning, across the concept, project and operational phases. On people, it augments existing roles and introduces new ones, covering how specialists such as data scientists and machine learning engineers fit alongside system and data owners, and the quality assurance unit's part in decisions. It also covers knowledge management and AI literacy, explainable AI, cybersecurity including adversarial attacks on data and models, and trustworthy AI themes such as human autonomy and control, transparency, accountability and bias mitigation.
What the public material does not tell you
ISPE's public material describes structure and topics, not detailed requirements. It does not set out the guide's validation framework, risk categories, test methods, documentation templates or performance thresholds. Neither ISPE's public topic lists nor the public table of contents name generative AI or large language models as a separate topic, although the product page describes the approach as suitable for all types of computerized systems using AI. The guide is industry good practice, not a regulation or regulator guidance; it sits alongside the regulations and guidance that apply to your records and products. Use this page to decide whether the guide is relevant to you, then work from the guide itself and your quality function's procedures.
Applying a risk-based framework starts with knowing where AI already touches regulated work. Atlas Bio's Secure AI for Life Sciences starts with an AI risk assessment of current AI use and data exposure. Status, October 2026. In use in Atlas Bio's own agent platform today: a separate verification step that checks agent work, and a hold that stops a task until a person decides. Built and tested on synthetic data, not yet used with a client: the AI risk assessment tooling (AI discovery, data-flow and vendor-terms mapping, and a NIST AI RMF gap report) and a record of what data an AI used, what it produced and who reviewed it. In development: a customer-controlled AI environment in your own Azure tenant, an AI gateway, agent permissions, evidence linkage and notifications. Each engagement states which controls are built. Decisions on GxP scope, validation and compliance stay with your quality function.
Does the GAMP AI guide replace GAMP 5?
No. ISPE describes it as a stand-alone guide designed to be used in parallel with GAMP 5 (Second Edition) and the other GAMP guides. GAMP 5 (Second Edition) already has an appendix on artificial intelligence and machine learning, D11, and the leads say the AI guide expands on its concept of dynamic systems.
Is the GAMP AI guide a regulatory requirement?
No. It is an industry good-practice guide from ISPE, not a regulation or a regulator's guidance. ISPE presents it as a risk-based framework to help AI-enabled computerized systems be fit for their intended use and compliant with applicable regulations. Meeting GxP expectations remains the regulated company's responsibility under the regulations and guidance that apply to it.
When was the ISPE GAMP AI guide published and how long is it?
ISPE lists it as published in July 2025, at 290 pages. It is a paid publication available from ISPE.
Does the guide apply to suppliers of AI software?
ISPE says it defines activities and responsibilities for both internal and external suppliers, and its table of contents has a chapter on supplier activities and an appendix on supplier management across all four life cycle phases. Suppliers, service providers and regulated companies are all named in its audience.
Does the guide cover AI that changes after deployment?
ISPE's leads say it expands on GAMP 5 (Second Edition) Appendix D11's concept of dynamic systems across the concept, project and operational phases, and the public table of contents lists AI sub-system adaptiveness under an AI maturity appendix. How the guide defines and controls such systems is in the guide itself.
- ISPE: GAMP Guide: Artificial Intelligence (product page, published July 2025)
- ISPE: GAMP Guide: Artificial Intelligence, public table of contents (PDF)
- ISPE iSpeak: New GAMP Guide Provides Framework to Achieve High-Quality AI-Enabled Computerized Systems for Life Sciences (15 August 2025)
- Pharmaceutical Engineering: New GAMP Guide Addresses Challenges Posed by AI-Enabled Computerized Systems (September/October 2025)
- Pharmaceutical Engineering: Message from the Chair - GAMP, An Evolving Framework (January/February 2026)
- ISPE: GAMP 5 - A Risk-Based Approach to Compliant GxP Computerized Systems (Second Edition, July 2022)
- ISPE: GAMP 5 (Second Edition) public table of contents, including Appendix D11 - Artificial Intelligence and Machine Learning (PDF)
Talk to Atlas Bio.
Sponsors, researchers and clinicians: tell us the decision you are trying to make and we will show what the platform can and cannot answer.
Contact Us →