TGA's evolving approach to AI as a medical device

The Therapeutic Goods Administration has signalled significant changes to how it classifies and regulates AI-powered software used in clinical settings. Under the current framework, many AI tools used in radiology fall into Class IIa or IIb medical device classifications, but the TGA is exploring whether the adaptive nature of machine learning models warrants a distinct regulatory pathway.

Central to the proposed changes is the concept of predetermined change control plans. Rather than requiring a new submission every time an AI model is updated, the TGA is considering allowing manufacturers to define an envelope of acceptable changes upfront. This mirrors the FDA's approach in the United States and would reduce the regulatory burden on vendors while maintaining safety oversight. However, it shifts more responsibility onto practices to verify that deployed versions match approved specifications.

For radiology practices, the practical impact is twofold. First, practices will need to maintain records of which software versions are deployed and confirm they fall within the manufacturer's approved change control plan. Second, practices using AI tools that are not TGA-registered, whether research prototypes or overseas products, face increasing regulatory risk. The TGA has indicated it will be more active in enforcement.

Attest helps practices navigate this by tracking tool registration status, software versions, and TGA classification details. When the regulatory landscape shifts, having a structured record of your AI governance decisions provides the evidence trail that protects your practice.