Australia Australia news media bargaining code

Australia's AI Copyright Fight Is the News Bargaining Code Again, and Opt-Out Is the Weaker Answer

A leaked opt-out proposal would let AI firms train on authored work unless creators object. Licensing with transparency would pay creators without slowing AI.

Australia's AI Copyright Choice People of Internet Research · Australia 86% Authors reporting income drop UK survey cited by the Australian … 2 Options in leaked paper Central-body payment or a quota of… 20% Small-publisher distribution loading Raised from 10% in the finalised N… peopleofinternet.com
Australia's AI Copyright Choice People of Internet Research · Australia 86% Authors reporting income drop 2 Options in leaked paper 20% Small-publisher distribution loa… peopleofinternet.com

Key Takeaways

Australian book publishers and authors told the Joint Select Committee on Artificial Intelligence last week that a leaked government proposal would turn copyright into an opt-out system. ABC News reported on 28 October that the sector describes itself as in 'crisis'. The committee's final report is due on 30 November. The question is the one Australia answered for news in 2021 and has kept revising since: do rights holders get paid through negotiated deals, or must they police the web and object after the fact?

What was actually proposed

The ABC's 15 September report on a consultation paper titled 'AI on Australian Terms' described two options put to rights-holder groups earlier that month. In the first, AI companies could train on 'unprotected' online material if they paid a central body, or licensed through Australian collecting groups. The second, from the Attorney-General's Department, would require AI companies to strike deals with a minimum number of companies for a minimum period before they got legal permission to train. Both would let owners opt out. One slide, as quoted by the ABC, said: 'No additional $ would be payable to creators or rights holders beyond deals negotiated to achieve quota.'

That is a weaker position than the one the government announced a year ago. Attorney-General Michelle Rowland ruled out a text and data mining exception in late October 2025, saying: 'This government has repeatedly said that there are no plans to weaken copyright protection when it comes to AI.' She also said the tech and creative sectors 'must now come together and find sensible and workable solutions'.

The best case for opt-out

The strongest argument for an opt-out is practical, and it deserves a fair statement. Frontier models train on billions of documents. Negotiating individual permissions at that scale may be impossible, and a legal regime that makes training in Australia unworkable could push model developers, and the local research and startup activity around them, offshore. Opt-out schemes are also a recognisable middle path, because they leave a veto with the owner. The government should take that concern seriously.

Why opt-out fails on its own terms

Opt-out puts the cost of compliance on the party least able to carry it. Hachette Australia and New Zealand head Louise Stark told the committee: 'AI companies seeking an opt-out model are asking government to reverse a basic principle: if you want to use someone else's work, you should first seek permission.' A mid-sized publisher or an individual author cannot track every crawler, mirror site and dataset. The ABC reports that authors already find AI-generated knock-offs of their books on sale, sometimes on publication day. Australian Publishers Association policy head Stuart Glover said instructional books such as travel guides and cookbooks face particular risk.

The income evidence is thin, and we should say so. The Australian Society of Authors cited a UK report finding that 86 per cent of authors surveyed reported a drop in income due to generative AI. That is a self-reported UK survey, not Australian market data. It shows the direction of concern, not the size of the harm. A pro-innovation analysis should not pretend the damage is proven, and it should not wave away a plausible harm either.

The larger flaw is that the leaked options, as reported, pay little. Under the quota model, creators receive only what the negotiated deals happen to deliver. A quota of deals with a handful of large rights holders could satisfy the legal test while most authors and small presses see nothing. That is permission without payment, with a compliance veneer.

What the news bargaining experience actually teaches

The news code is the obvious precedent, and it cuts in both directions. Treasury's 3 August 2026 release finalising the News Bargaining Incentive shows how much the model has had to change. The government is 'narrowing the charge base while increasing the rate of the incentive'. It is lifting the distribution loading for regional, small to medium, and underrepresented-community publishers from 10 to 20 per cent. It is also broadening the journalist definition to include freelancers 'producing core new content'.

Two lessons follow. First, bargaining regimes drift toward administrative complexity, because the first round of deals favours the biggest publishers and each fix adds machinery. Second, the fix is an explicit distribution rule for small players, which a pure quota-of-deals model does not have. If Australia reaches for a bargaining design for AI, it should build the distribution rule in from the start, rather than discovering the gap in year five.

A proportionate alternative

A workable package has three parts, none of which requires a blanket exception or a blanket ban.

The government has already said no to a text and data mining exception. Reversing that by the back door, through an opt-out whose practical effect is free training, would damage trust with the creative sector. It would also give AI developers legal certainty that lasts only until the first court challenge. Licensing is slower to set up, but it produces durable certainty and a domestic market for training data.

What to watch on 30 November

The committee has heard from rights holders, broadcasters and AI developers. Its report can recommend a path without settling the details. The test is whether it endorses payment and transparency as the baseline, or treats opt-out as an acceptable compromise. Australia's experience with news says the details of who gets paid, and how, decide whether a scheme works. The headline design matters less.

Sources & Citations

  1. ABC News: Australian book publishing in 'crisis' over rise of AI (28 Oct 2026)
  2. ABC News: Leaked government copyright proposal (15 Sep 2026)
  3. Treasury ministers: News Bargaining Incentive legislation finalised (3 Aug 2026)
  4. ABC News: Government rules out AI training copyright exception (27 Oct 2025)
  5. iTnews: Government rules out AI training exemption in copyright laws