This submission to the Proposal Paper on introducing mandatory guardrails for AI in high-risk settings supports a risk-based approach to ensure AI is developed and used safely and responsibly in Australia, particularly in high-risk applications such as healthcare (including medical devices), transportation, law enforcement, and financial services. 

The authors recommend a framework approach to existing legislation with amendments in the near term, with mandatory guardrails for AI in high-risk settings. It is recognised that several countries leading AI regulation reforms are adopting a risk-based approach focusing on ex-ante (preventative) measures, which is laudable. However, given the complexity and pace of AI technology development and uptake, a regulatory regime that anticipates preventative intervention or mandating standards will be insufficient. Preventative intervention and mandating standards cannot fully satisfy the objective and substantiated concerns of algorithmic bias and discrimination, lack of transparency and accountability, cultural and data sovereignty erosion, digital exclusion and access gaps, and surveillance and privacy concerns. In the following response, we provide feedback and justify our evidence-informed recommendations.

The submission's recommendations aim to align with fairness, transparency, and human rights principles, which must be articulated and regularly reinforced within the government as an exemplar to industry. The recommendations are designed to be incorporated within various mechanisms, including principles, frameworks, regulations, amendments, and future AI legislation.