Australian AI Governance Briefing: Week Ending 20 September 2026
A leaked Attorney-General's Department options paper setting out an opt-out model for AI training on Australian creative works was tabled in the Senate, as PM&C opened consultation on mandatory national AI infrastructure standards.
Copyright and AI training dominated the week. On Tuesday 15 September ACT independent Senator David Pocock tabled in the Senate a leaked Attorney-General’s Department document titled “AI on Australian terms”, setting out copyright reform options under which AI companies could train on Australian creative works unless creators opt out. Attorney-General Michelle Rowland’s office said the government was consulting creators, media organisations and AI companies on a range of options. OpenAI said the same day that current copyright settings stood in the way of it building an AI training centre in Australia.
On Friday 18 September the Department of the Prime Minister and Cabinet opened public consultation on “Getting it right: Building AI infrastructure that works for Australia”, covering mandatory national standards for data centre energy, water and land use and conditions for AI training, which the government plans to legislate. Submissions close at 5pm AEDT on 9 October 2026. Copyright is outside the consultation’s scope.
In Hobart, the Supreme Court of Tasmania on 15 September ruled invalid a parole condition after hearing the Parole Board had relied on a document citing case law that did not exist and was likely AI-generated. Tasmania’s Department of Justice has since confirmed a review of AI use in past parole decisions. The Director-General of the Australian Signals Directorate also called for an AI “early warning system”.
The week in review
Leaked Attorney-General’s Department paper sets out opt-out options for AI training on creative works
On Tuesday 15 September 2026 ACT independent Senator David Pocock tabled in the Senate a leaked Attorney-General’s Department document titled “AI on Australian terms”. The confidential document sets out copyright reform options for the use of Australian creative works to train AI models, and had been presented to rights-holder groups in consultations. Excerpts were first reported by the ABC.
Under one option, AI companies that strike deals with a minimum number of rights holders for a minimum period would be permitted to use content for AI training. Creators would be able to opt out to shield their work; for those who remain in, any payment would depend on the agreements negotiated with AI companies. A second option would provide compensation for all scraped work while also allowing creators to opt out. Under either option the default would move from creators granting permission to creators having to object.
Senator Pocock said the opt-out approach would upend copyright as Australians know it and described it as “throwing creatives under the bus”. Opposition spokeswoman Michaelia Cash said existing copyright law is the correct framework and should not be watered down for AI companies. Greens senator Sarah Hanson-Young also criticised the options. A spokesperson for Attorney-General Michelle Rowland said the government was consulting creators, media organisations and AI companies on a range of options for copyright changes.
The Prime Minister said in a July 2026 speech announcing a national AI framework that no company should use Australian books, music, art or news to train AI without the artist’s control, including control of the price and value of their work.
On the same Tuesday, OpenAI said it would not establish AI training centres in Australia, or invest in renewable energy to support them, unless copyright settings changed. Nationals leader Matt Canavan and Senator Hanson-Young described the comments as blackmail. Deputy Prime Minister Richard Marles told the ABC’s 7.30 that the economic opportunity in collaborating with frontier AI companies was enormous, and that it needed to come on terms in the national interest.
Primary sources: AAP — Labor floats idea to let AI use artists’ work for free | PS News — Labor considering options to water down copyright laws in favour of global AI titans
Tasmanian Supreme Court voids parole condition after board relied on AI-hallucinated case law
On Tuesday 15 September 2026 Justice Stephen Estcourt of the Supreme Court of Tasmania ruled invalid a Parole Board of Tasmania condition preventing Susan Neill-Fraser from speaking to the media to claim her innocence or wrongful conviction. His Honour found the condition denied her procedural fairness, was legally unreasonable and created uncertainty. The court also accepted that the board’s decision-making had been affected by a document citing case law that did not exist and was likely the product of “AI hallucinations”.
Neill-Fraser’s counsel, Patrick Coleridge, told the court the parole board secretary had used AI to prepare a rationale supporting the condition. Neill-Fraser’s legal team raised the non-existent case law with the court in August, after which the board met and withdrew the condition that month. The parties had agreed the condition should be voided, but the court’s jurisdiction was required for that outcome to take effect. The condition was first imposed in December 2024 and replaced in May 2025.
Tasmanian Attorney-General Guy Barnett said he would write to the Parole Board seeking an explanation of its professional standards, stating that “the production of inaccurate information by a public authority is unacceptable”. Opposition parties called for a review of past cases. On 18 September, in a statement reported by the ABC on 19 September, Tasmania’s Department of Justice confirmed it would review the extent to which AI may have been used to inform past Parole Board decisions. Human Rights Law Centre legal director Sarah Schwartz said the case exposed serious deficiencies in Tasmania’s parole system.
Primary sources: ABC News — ‘AI hallucinations’ in parole board submission scuttles media ban (15 September) | ABC News — Attorney-general demands explanation from Tasmanian parole board (16 September) | ABC News — Neill-Fraser revelations lead to review of AI use in parole decisions (19 September)
PM&C opens consultation on mandatory national AI infrastructure standards
The Office of AI within the Department of the Prime Minister and Cabinet opened public consultation on “Getting it right: Building AI infrastructure that works for Australia” at 6am AEST on Friday 18 September 2026. Submissions close at 5pm AEDT on Friday 9 October 2026 and are made through an online form, with a list of consultation questions provided.
The Australian Government will work with states and territories to create mandatory standards for how data centres use energy, water and land, and for how AI training can support skills and industry development. The standards are intended to operate alongside state and territory planning and approval processes, and the government plans to legislate them. The consultation paper states that the standards will build on the national Expectations of data centres and AI infrastructure developers released in March 2026 and align with National Cabinet’s decision in August 2026.
The consultation also seeks views on whether current policies, regulations and infrastructure can support AI training in Australia. PM&C states that copyright policy is outside the scope of the consultation because of work already underway, with the Attorney-General continuing to lead on copyright. Submissions will be considered alongside targeted stakeholder engagement and ongoing work with Commonwealth, state and territory governments.
Primary sources: PM&C — Getting it right: Building AI infrastructure that works for Australia | PM&C — Consultation paper, September 2026 | PM&C — Have your say on the future of AI training and infrastructure in Australia
ASD Director-General calls for an AI early warning system
Speaking at the Australian Strategic Policy Institute’s Sydney Dialogue AI Masterclass in Canberra on Monday 14 September 2026, Australian Signals Directorate Director-General Abigail Bradshaw said Australia has an early warning system in a cyber context but does not yet have one formalised for AI. She said intelligence and security agencies need to work with AI companies to identify early risks.
Ms Bradshaw said ASD had used access to advanced AI models to help Australian organisations fortify their systems and identify vulnerabilities, completing in hours work that would have taken ASD cybersecurity experts weeks. She said her agency could not reliably count the number of AI agents active on the internet, and that governments and businesses would face enormous costs to update outdated technology. She urged organisations to maintain basic security measures while using AI to defend their networks, and called for national security agencies to be involved in the independent model evaluations proposed by Anthropic.
At the same event Assistant Minister Andrew Charlton said he did not favour leaving AI guardrails to the companies or to the superpowers.
Primary sources: ABC News — Australia needs an AI ‘early warning system’, top cybersecurity chief warns | PS News — Warnings over AI as government wants to attract more investment
Stories
Leaked Attorney-General’s Department paper proposing opt-out AI training model tabled in the Senate
Senator David Pocock tabled a leaked Attorney-General’s Department document titled “AI on Australian terms” in the Senate on 15 September 2026. The document sets out copyright reform options under which AI companies could train on Australian creative works unless creators opt out, including one option where payment would depend entirely on deals struck with a minimum number of rights holders. The Attorney-General’s office said the government was consulting creators, media organisations and AI companies on a range of options.
Tasmanian Supreme Court voids parole condition after board relied on AI-hallucinated case law
On 15 September 2026 Justice Stephen Estcourt ruled invalid a Parole Board of Tasmania condition barring Susan Neill-Fraser from speaking to the media, finding it denied procedural fairness, was legally unreasonable and created uncertainty. The court accepted that the board’s decision-making was affected by a document citing non-existent case law that was likely AI-generated. Tasmania’s Department of Justice has confirmed it will review the extent to which AI was used to inform past Parole Board decisions.
PM&C opens consultation on mandatory national standards for data centres and AI training
The Department of the Prime Minister and Cabinet opened consultation on “Getting it right: Building AI infrastructure that works for Australia” on 18 September 2026. The government will work with states and territories on mandatory standards for data centre energy, water and land use and for how AI training supports skills and industry development, and plans to legislate them. Submissions close at 5pm AEDT on 9 October 2026, and copyright policy is outside the consultation’s scope.
ASD Director-General calls for an AI early warning system
Australian Signals Directorate Director-General Abigail Bradshaw told the ASPI Sydney Dialogue AI Masterclass on 14 September 2026 that Australia needs a formalised AI early warning system equivalent to the one it has for cyber threats. She said intelligence and security agencies need to work with AI companies to identify early risks, and urged organisations to use AI defensively while maintaining basic security measures.
This briefing was researched and written with AI assistance.
Stay across Australian AI governance
Get the briefing delivered to your inbox every week. No spam, unsubscribe any time.