Good AI Practice Principles 8–10: Assurance

The final three principles focus on proving AI works as intended, maintaining performance over time, and communicating honestly with those who rely on AI outputs. These are especially relevant for pharmacovigilance teams, clinical trial monitors, and regulatory submitters.

Principle 8: Risk-based performance assessment

Risk-based performance assessments evaluate the complete system including human–AI interactions, using fit-for-use data and metrics appropriate for the intended context of use.

Validation of predictive performance is supported by appropriately designed testing and evaluation methods.

Why human–AI interaction matters

In drug development, AI rarely operates in isolation. A pharmacologist reviewing AI-flagged signals, or a clinician interpreting trial enrichment recommendations, forms part of the complete system. Performance assessment must reflect real workflow — not laboratory conditions alone.

Questions for Hong Kong teams

• Were metrics chosen for your context of use?
• Does validation include human–AI team performance?
• Is test data fit-for-use and independent of training?

Principle 9: Life cycle management

Risk-based quality management systems are implemented throughout the AI technologies' life cycles, including to support capturing, assessing, and addressing issues.

AI technologies undergo scheduled monitoring and periodic re-evaluation to ensure adequate performance — for example, to address data drift.

Data drift occurs when real-world data gradually differ from training data, potentially degrading model performance. This is a growing concern as AI supports ongoing pharmacovigilance and manufacturing analytics.

Shared responsibility

Sponsors and vendors must build monitoring processes; local sites and hospital departments must define who reviews alerts, when to escalate, and when to pause use pending investigation.

Principle 10: Clear, essential information

Plain language is used to present clear, accessible, and contextually relevant information to the intended audience — including users and patients — regarding:

• The AI technology's context of use
• Performance and limitations
• Underlying data
• Updates
• Interpretability or explainability

For clinicians and pharmacists

Before relying on AI-generated evidence in a submission or clinical decision support workflow, confirm you have received understandable documentation — not only technical appendices written for data scientists.

Responsible use means knowing what the model cannot do, not only its headline performance.

Source: FDA & EMA — Guiding Principles of Good AI Practice in Drug Development (January 2026)