Labs use AI, for example, to flag anomalous chromatograms and draft documents. You must be able to explain and control what the system does. LBP assesses AI applications and documents the controls; the model itself comes from you or your supplier. By smart systems we mean rules-based automation and decision-support systems.

Deliverables

Use-case and suitability assessment
Intended-use specification and risk assessment
Data assessment and test strategy (independent test data never used in training)
Validation report with acceptance criteria and known limitations
Monitoring, change control, incident and fallback procedure

Four kinds of systems, each with its own assurance

We distinguish three kinds of AI (locked ML models, adaptive or retrained models, and generative AI) plus rules-based automation. Each type needs its own evidence. For example, a model flagging anomalous chromatograms needs different controls than a language model drafting SOPs. Even when a person reviews every output, we set validation depth by risk. Assessor competence and independence also matter, as does the risk that people follow the system blindly.

Regulatory status as of 2 October 2026. In July 2025 a draft of EU GMP Annex 22 on artificial intelligence was published for consultation; that draft describes only locked, deterministic models for critical GMP applications. Article 4 of the AI Act (AI literacy) has applied since 2 February 2025. Medical devices and IVDs face high-risk AI requirements from August 2027. We also follow the ISPE GAMP guide for AI (2025), the EMA reflection paper on AI in the medicinal product lifecycle and the FDA draft guidance of January 2025.

AI validation for GxP laboratories

Deliverables

Four kinds of systems, each with its own assurance

Ready to talk?