The APS AI use policy and mandatory obligations
The Australian Public Service Commission's AI use policy establishes a framework for Commonwealth agency AI deployment that goes beyond voluntary guidance. The policy's core principles — accountability, human oversight, transparency, fairness, privacy, and reliability — create specific operational expectations for agencies deploying AI. Critically, the policy distinguishes between AI that assists human decisions (which is generally permissible with appropriate governance) and AI that makes autonomous decisions about individuals (which requires specific legal authority and more demanding governance).
The accountability principle is the most demanding: agencies must be able to explain and justify decisions made with AI assistance. This requires that agencies maintain records of AI system inputs and outputs sufficient to reconstruct the basis for any individual decision, that human decision-makers genuinely review AI recommendations rather than rubber-stamping them, and that accountability for outcomes rests with identifiable human officials, not with the AI system.
Administrative law and algorithmic decision-making
Commonwealth administrative law — the Administrative Decisions (Judicial Review) Act, the Acts Interpretation Act, and the common law of administrative law — creates a framework for government decision-making that predates AI by decades but applies to it with full force. Decisions made by Commonwealth agencies must be authorised by law, must be procedurally fair (which generally means giving affected parties an opportunity to be heard), must be rationally connected to the evidence, and must be made by the decision-maker authorised to make them.
Each of these requirements creates specific obligations for AI-assisted government decision-making. The authorisation requirement means that an AI system cannot make decisions that the authorising legislation gives to a human official — the human must genuinely decide, not simply endorse an automated recommendation. The procedural fairness requirement means that individuals who may be adversely affected by an AI-influenced decision must have an opportunity to present their case to a decision-maker with genuine discretion. And the reason-giving obligations that apply to many government decisions must be satisfied by explanations that genuinely explain the basis for the decision — reference to an algorithm or a risk score is not adequate.
Services Australia and welfare AI: the Robodebt legacy
Services Australia's AI governance is shaped directly by the Robodebt Royal Commission's findings. The Commission found that automated debt generation without legal authority was unlawful, that the reversal of the burden of proof was unjust, and that the deliberate suppression of documentation was a serious breach of public service standards. Services Australia has since implemented significant reforms to its automated decision-making governance, but the Commission's findings continue to inform Commonwealth agency AI governance more broadly. The key lesson for all government agencies: the combination of automated decision-making, vulnerable affected populations, and inadequate oversight creates the conditions for Robodebt-type failures. Each element must be actively governed.
