TLDR: Australia’s Therapeutic Goods Administration (TGA) has identified an urgent need for comprehensive guidance on adaptive and generative artificial intelligence (AI) within medical devices. While the existing regulatory framework is largely considered robust, stakeholders are calling for clearer guidelines on the oversight, validation, and accountability of these rapidly evolving AI technologies.
The Therapeutic Goods Administration (TGA) in Australia has underscored a critical requirement for new guidance addressing adaptive and generative artificial intelligence (AI) models in medical devices. This pressing need emerged from a public consultation conducted by the TGA, as reported by the Regulatory Affairs Professionals Society (RAPS) on August 4, 2025. The consultation, part of the Australian Government’s ‘Supporting Safe and Responsible AI Budget’ measure, aimed to clarify and strengthen the regulation of medical device software, including AI.
Despite over 90% of respondents affirming the robustness and flexibility of the TGA’s current technology-agnostic framework for regulating AI, the consultation highlighted specific areas demanding immediate attention. Stakeholders, particularly from the medical device industry, expressed a strong desire for clarity on how to manage and account for the outputs of adaptive or generative AI models, especially when the original deployer lacks oversight of the relevant output.
A significant concern raised was the continuous adaptation of AI, which stakeholders believe necessitates constant monitoring and real-time, continual review of performance to prevent degradation of output quality over time. The TGA acknowledged the impracticality of such continuous human oversight, noting that ‘These approaches are likely to be impractical without involving further AI models and systems tasked with conducting rolling review.’
In response to calls for more information on regulatory processes for assessing and approving changes in adaptive models, the TGA stated that guidance for technical requirements for adaptive and generative AI should be developed ‘as expeditiously as possible.’ Furthermore, respondents urged the TGA to provide more detailed information on how essential safety and performance principles apply to the ongoing validation of these advanced AI systems.
The TGA has opted, for now, against altering its existing essential principles. Instead, it plans to develop more extensive guidance, incorporating examples to demonstrate how current regulations apply to specific use cases. However, the agency indicated that more fundamental changes to essential principles and conformity assessment procedures might be considered as the complexities of regulating adaptive and generative AI continue to evolve.
An emerging challenge identified by the TGA is the development of innovative devices trained on unvalidated open datasets. Developers are seeking clearer information on acceptable validation methods for medical devices incorporating software of unknown provenance or trained on such open datasets.
Also Read:
- Artificial Intelligence and Generative AI Driving Transformative Growth in Healthcare Sector
- Governments Worldwide Accelerate AI Integration in Public Services
Additionally, 81% of stakeholders agreed that a review of definitions within the Therapeutic Goods Act and subordinate legislation is necessary to clarify responsibilities for the development, deployment, and use of AI models and systems. Concerns were voiced that existing legislation does not clearly stipulate accountability for the outcomes of advanced AI technologies, particularly when outputs could constitute an offense under the Act. Health professionals, in particular, strongly advocated for legal responsibility for the safety, quality, and performance of AI-driven software to be assigned to manufacturers and sponsors.
