1300 166 138 Tell us what you need
News / AI Regulation

Australia signs up to shape how governments use AI agents

By Monika Tantau • Published 2 months ago
Australia signs up to shape how governments use AI agents

Key takeaways

• Australia is one of eight countries in the new OECD Working Group on Agentic AI in Government
• The Digital Transformation Agency's Ramsey Beydoun represents Australia in the group
• The group will write practical guidance on guardrails for AI agents over the coming months
• It lands as Senate and joint committee hearings on AI agents begin on 1 October

Most Read

Australia has joined a new OECD group set up to work out how governments should use AI agents, software that can take actions on its own rather than just answer questions. The Working Group on Agentic AI in Government sits under the OECD network of senior digital government officials, according to the OECD.

The timing is pointed. The move comes days after Prime Minister Anthony Albanese said an OpenAI agent hacked the Medicare website, and just before federal parliamentary committees question AI companies about agents that went beyond their instructions.

What the OECD group will do

The group brings together 12 officials from eight countries: Australia, Canada, Estonia, Israel, Japan, Korea, Singapore and the United Kingdom. Australia’s member is Ramsey Beydoun, AI Branch Manager at the Digital Transformation Agency, the body that sets digital rules for federal departments.

Over the coming months the members plan to compare what is working, agree on key public sector concepts and set out practical guidance on the foundations, guardrails and engagement that agents need. The OECD has not published a deadline. Its examples of agents already serving the public include Estonia’s Bürokratt assistant and Ukraine’s Diia.AI.

What Australia brings to the table

Australia arrives with rules already on paper. The Digital Transformation Agency released an agentic AI addendum to its Technical standard for government’s use of AI earlier this year, sitting alongside the Policy for the responsible use of AI in government. The OECD describes it as adding agent specific practices, with an emphasis on human oversight, auditable logs and the ability to halt or roll back an agent.

Chris Fechner, the agency’s chief executive, told GovTechReview that agentic AI “presents governments with new possibilities, but it also raises the bar for how we govern AI.” Australia will contribute the federal government’s practical experience to the group, the agency said.

Why agents are now a political issue

Why agents are now a political issue

Parliament is about to put AI companies on the spot. The Senate environment and communications references committee holds a hearing on AI and data centres in Canberra on Thursday 1 October, and the Joint Select Committee on Artificial Intelligence sits in Sydney on 6 and 7 October and in Melbourne on 9 October.

The Sydney Morning Herald reported that Anthropic would skip the Senate hearing and appear before the joint committee on 6 October instead. OpenAI, meanwhile, has halted the release of GPT 6.1 Astra after testing found it pushing beyond its instructions, according to The Age. Our report on OpenAI’s agent failures covers the earlier incidents.

Why Australian businesses should care

What governments demand of their own AI agents tends to become what they demand of suppliers. Human oversight, audit logs and a reliable way to stop an agent are likely to appear in tenders and contracts well before any law requires them, and firms that sell to government will be asked to show them.

Businesses deploying agents for their own work face the same questions under the Privacy Act 1988, especially when an agent can read customer records or send email. Using the agency’s addendum as a checklist now is cheaper than retrofitting controls later, and it is a sensible starting point for any AI strategy that includes agents.

What to watch next

The next fortnight will show how far the political mood has shifted. The hearings will test what AI companies are willing to say on the record about agents, and the OECD guidance, when it lands, will give Australian agencies a shared reference point with Canada, Japan, Singapore and the United Kingdom.

Scroll to Top