Government AI decision-making: Australia plans new curbs

Australia is drawing up rules to limit how federal departments and agencies use automated AI when making decisions, putting a concrete compliance measure into Labor’s national AI plan. The Guardian first reported that the proposed standard would curb automated decision-making inside government, moving Canberra’s AI debate from promotion to the systems that can affect citizens.

For departments, that turns a strategy document into a rule book for their own tools.

The proposed standard sits beside a wider push for digital duty of care laws covering consumer protection, workplace safety and privacy, according to ABC’s reporting on the plan. The package suggests Labor wants AI governance to reach past voluntary principles. It is also a more awkward question than attracting compute capacity or datacentre investment: how ministers expect agencies to use software when administrative decisions are at stake, and what happens when a system’s recommendation is wrong.

Assistant minister for science, technology and the digital economy Andrew Charlton cast the policy as an enforceability issue, not just another argument for adoption. In the prime minister’s plan, he said public acceptance would depend on clearer limits around AI use: “Establishing a clear and enforceable social licence for AI is fundamental towards achieving this objective.”

Prime Minister Anthony Albanese told ABC News earlier in July that “AI touches on the work of every minister and department”. He said stronger guardrails would impose “an important discipline on government”. Those comments put public-sector deployment near the centre of the plan, rather than treating it as a side issue for procurement teams.

The timeline is tightening. The government has flagged an Office of AI, with National Cabinet expected to consider the broader framework in August 2026 and legislation targeted for early 2027. That schedule matters because agency procurement cycles are slow; if the framework holds, some departments will need to start mapping automated decision points before legislation is introduced.

For agencies, any curb on automated decision-making would shift the work from guidance to process. Departments would need to decide where software can assist, where human review has to stay in the loop, and how risk is recorded when a tool affects the public. That is more operational than the growth agenda that has dominated much of Australia’s recent AI discussion. It also gives vendors less room to sell generic automation without explaining how their systems are checked inside government workflows.

The work is landing as ministers deal with rapid growth in AI use and a datacentre construction boom, another theme in The Guardian’s reporting. The government standard is different in character. Instead of selling AI as an economic opportunity, it asks how Canberra will constrain its own use of the technology when accountability and administrative fairness are on the line. The public documents released so far set direction more than detail, and the Guardian and ABC reports show ministers are still drafting the standard. Agencies, vendors and affected citizens are left waiting for the detail, but the planned curb is already a clearer signal than the usual AI principles language.