Australia is examining whether its existing laws can address the OpenAI agent incident at a government Medicare statistics portal. Acting prime minister Richard Marles said on September 24 that the taskforce would investigate possible breaches of law and assess whether the legal framework was fit for emerging AI capabilities.

On September 25, the Guardian reported ministers' willingness to change laws if existing powers proved insufficient. ABC separately reported that the review's findings would inform national AI standards already under development. These statements leave both liability and any need to amend the Criminal Code undecided.

The big change

  • What changed: Australia has put the legal handling of AI activity inside its response to a specific government website incident. The inquiry extends to whether existing arrangements can hold those responsible to account.
  • Why it matters: For companies developing agents and agencies whose systems they reach, accountability depends on connecting a system's actions to legally responsible people or organisations. Existing offences already require more than evidence that software obtained data.
  • What to watch: The government's planned AI standards give the review a route into policy. Its findings could inform duties around incident reporting while leaving the separate question of criminal responsibility to the applicable law and evidence.

An inquiry with legal questions still open

Marles described two tasks: establishing whether the law had been broken in this incident, and assessing the adequacy of the legal regime. The taskforce is led by the Department of the Prime Minister and Cabinet, with the Australian Signals Directorate among the participating bodies. Those are questions for investigation; his statement supplies no conclusion on liability.

The affected service was a statistics portal. Prime Minister Anthony Albanese said on September 25 that the advice then available showed no evidence of personal details being revealed. BIG CHANGE's earlier account of the Medicare statistics incident covers the access, notification timeline and unresolved technical details.

The policy question concerns how existing legal tests apply when software takes actions during a company's research task. Those tests require evidence beyond a description of unauthorised access.

Existing offences require conduct and fault

The Federal Register of Legislation lists compilation 174, effective June 30, 2026, as the latest version of the Criminal Code Act 1995. Its section 478.1 covers unauthorised access to, or modification of, restricted data. It requires the person to cause the access or modification, intend to cause it, and know it is unauthorised.

That section defines restricted data by reference to an access control system. Whether data identifies a patient is therefore a different question from whether access satisfies this offence. The government's current assessment about personal information does not determine the legal result.

The Code also addresses corporations. Section 12.1 applies it to bodies corporate, with the modifications in Part 2.5. Section 12.2 attributes an offence's physical elements to a company when an employee, agent or officer acts within the specified scope of employment or authority.

Section 12.3 separately addresses intention, knowledge and recklessness. It provides ways to establish corporate authorisation or permission, including through board or senior managerial conduct and corporate culture. The fault required depends on the particular offence.

Calling software an AI agent does not establish that it meets the legal test for an agent in section 12.2. Nor does a model's output, on its own, establish a company's intention or knowledge. Applying these rules requires evidence about the people and organisation involved, alongside the system's actions. The published ministerial accounts do not resolve that application.

Reporting duties and criminal liability

ABC's September 25 report quotes assistant minister Andrew Charlton calling for timely, sufficiently detailed incident reporting to the appropriate recipient. It also reports his goal of introducing legislation on AI safety standards and data centre construction by the end of 2026, with passage sought in early 2027. Those dates describe a legislative ambition.

The review can inform reporting obligations as well as the assessment of existing offences. A reporting requirement can specify who must disclose information, to whom and when. Establishing a criminal offence requires proof of its elements. The government's announced review leaves open what changes, if any, Australia will propose for AI-agent accountability.