The copyright papers leaked, and both options make Australian creators say no in advance.
Curated by Roger Hanney · Throughline Advisory · Sydney · Edition of Thursday 17 September 2026, covering 14 to 17 September
Published 22 September 2026
Ask first is the current law. Both leaked options replace it with tell us if you object.
The government ruled out a blanket training exemption last December. The papers show the two alternatives it is weighing.
Under Australian copyright law an AI company must obtain permission before copying a work to train a model. There is no general exception for text and data mining. On 15 September the ABC published leaked consultation documents from confidential talks the Attorney-General's Department held earlier in the month with creators, media companies and AI firms.
A spokesperson for Attorney-General Michelle Rowland told the ABC the options are intended to give creators control and fair payment. On 16 September the ABC reported that Jeffrey Bleich, Anthropic's special international envoy, and Ann O'Leary, OpenAI's vice president of global policy, had met senior ministers earlier in the week. OpenAI's written submission to the parliamentary AI inquiry says Australia's copyright settings are part of the investment environment.
The objections came from every side of the chamber. Shadow attorney-general Michaelia Cash and communications spokeswoman Sarah Henderson called for the full consultation material to be released and for training on copyright material without consent to be ruled out. Greens arts spokeswoman Sarah Hanson-Young said AI companies should pay for Australian content like everyone else. Holly Rankin, who records as Jack River, said the proposals shift the burden of protecting copyright onto individuals and break assurances that creators would keep control.
On 17 September The Conversation published a legal analysis arguing that an opt-out model does not fit Australian copyright law. The law grants exclusive rights to the owner. It does not grant conditional rights to the user. That is the argument the Senate will hear.
Sources: ABC News, 15 Sep 2026, ABC News, 16 Sep 2026 and The Conversation, 17 Sep 2026.
Who has to act, under the current law and under each leaked option.
| Current law | Option 1: minimum deals | Option 2: central fund | |
|---|---|---|---|
| Who moves first | The AI company asks | The creator opts out | The creator opts out |
| Who gets paid | Whoever licenses | Rights holders inside the minimum deals | Rights holders, via the fund |
| Who is not paid | Nobody is copied without a licence | Creators outside the deals | Depends on the fund's distribution rules |
| What a silent creator gets | Nothing is taken | Work is trained on, unpaid | Work is trained on, paid via the fund |
Every organisation that publishes holds copyright. Under either option the default for your manuals, research, images and marketing becomes available for training unless someone in your organisation has said otherwise. Silence changes meaning.
Sydney's two largest universities signed with the same vendor in the same week.
Issue 018 covered UNSW. Four days later its neighbour did the same, with a different integrity model.
On 15 September iTnews reported that the University of Sydney is rolling out ChatGPT Edu to 80,000 students and staff. Features will be switched on in stages under the university's responsible AI framework. Integrity is handled by its existing two-lane assessment strategy: some assessments are secured against AI use, the rest assume it. The university is also spending $1 million on research into AI's environmental impact.
Two institutions, 160,000 people, one supplier, one week. UNSW chose three labels per assessment; Sydney chose two lanes. Both concluded that detection is not a strategy and that the rule has to be set before the work is done. Both also chose the same vendor, which is a concentration worth noting when OpenAI is lobbying Canberra on copyright in the same fortnight.
Source: iTnews, 15 Sep 2026.
Opt-out with payment is the only version of copyright that has ever paid the long tail.
The lead treats the reversal of burden as a loss for creators. The counter-case says the current rule pays almost nobody. Ask-first works for a record label with a licensing desk. It does nothing for a poet with one book, because no AI company will find her, negotiate with her and pay her. Under the current law her work is simply not used, and she receives nothing. Under Option 2 her work is used and a fund sends her a cheque. Collecting societies already run on this model for radio and photocopying, and Australian creators have banked those payments for forty years.
The counter-case also names the alternative. If Australia keeps ask-first while the United States argues fair use, training happens offshore on the same Australian works, and the poet still gets nothing. The choice is not between payment and no payment. It is between payment through a fund and no payment at all.
"An Act for the Encouragement of Learning, by vesting the Copies of printed Books in the Authors or Purchasers of such Copies."
Register your own copyright works before a law asks you to.
Use case. Anyone whose organisation publishes. Australia has no copyright registration system; the right arises automatically. That means most organisations have no list of what they own. Make one. Take each class of work: manuals, reports, photography, code, marketing, training material. Record where it is published and who created it. Note whether the copyright is yours or a contractor's. Then mark it license, refuse or undecided. A first pass takes two hours.
Tips. Check contractor and agency agreements first; assignment clauses are often missing. Note which works are already behind a login and which are public, because the leaked options only bite on what can be reached. Pair the register with the robots.txt rules from Issue 020 so the technical signal matches the legal position.
Learn more, free, no paywall: the ABC's account of the two options for what the register would be used for. The Conversation's legal analysis explains how the current law allocates the right.
Write one sentence stating your organisation's position on AI training, and put it where a crawler and a lawyer would both find it.
Either option would make silence into consent. A sentence on your website's terms page, matched by the crawler rules in your robots.txt, is the cheapest opt-out you will ever file, and it is dated. Decide the sentence now, before the consultation closes and the default moves. Drafting that position with a commercial lens is the engagement Throughline Advisory runs: throughlineadvisory.au.