AI and Model Governance Move from Theory to Regulatory Expectation

The TGA opened consultation on 26 October on whether to keep, remove or modify the medical-device exclusion for certain digital mental-health tools, citing increasingly complex apps and chatbots that use AI and personalised recommendations. Removing or narrowing the exclusion could bring these products under conformity assessment, clinical evidence, software lifecycle and post-market monitoring requirements, and the consultation closes on 7 December 2026. In the US, FDA held a stakeholder webinar on 29 October to help companies put ICH M15 on model-informed drug development into practice; the guideline, final since June 2026, sets shared principles for planning, evaluating and documenting models that inform regulatory decisions.

Regulators are converging on one principle: a model or algorithm must be assessed against its context of use and the decision its output supports, not simply declared “validated”. Model governance can no longer sit only with data-science or pharmacometrics teams; it belongs inside the quality system.

Set up one controlled register covering GxP AI, statistical models and algorithmic decision-support tools, recording each tool’s intended use, decision impact, version control and change triggers.

https://www.tga.gov.au/products/regulations-all-products/tga-reforms/medical-devices-reforms/medical-devices-reforms-software-regulation/review-exclusion-certain-digital-mental-health-tools

Speak to BethAri about building a risk-based AI and model governance framework.