Skip to content

Australia's AI Copyright Fight: Who Owns the Value of Training Data?

ArticleBy Factrail

By the · Method: how we source and score · Last reviewed

The State Library of New South Wales on Shakespeare Place, Sydney

TL;DR

Australian politicians clash over AI training and copyright: creators' compensation, tech investment and energy costs. Reporting published over 14–15 November 2026 described South Australian senator Andrew McLachlan's criticism of a rush to relax copyright rules for the benefit of technology companies.

Key graph impact

The article explains graph entities; the causal edges belong to the facts and drivers it links to.

  • Strengthens · 70% impact strength

    Observed

    29.8% of population aged 16-65 years · 8.13 below the normal line (2023)

    Worse than normal

    Two readings: 30.2% of population aged 16-65 years (2021) and 29.8% of population aged 16-65 years (2023).

    20212023

    Above normal = better · Lag ~6 mo

  • Strengthens · 62% impact strength

    Observed

    10.4% · 10.2 above the normal line (2024)

    Worse than normal

    7 readings, 2018–2024: fell from 11.1% to 10.4%.

    Above normal = worse

  • Weakens · 60% impact strength

    Observed

    7.97% of population aged 16 years or over · 5.2 above the normal line (2025)

    Worse than normal

    8 readings, 2018–2025: rose from 7.77% of population aged 16 years or over to 7.97% of population aged 16 years or over.

    Above normal = worse · Lag ~3 mo

  • Strengthens · 50% impact strength

    Observed

    91.6% of population · 8.4 below the normal line (2023)

    Worse than normal

    4 readings, 2020–2023: rose from 90.4% of population to 91.6% of population.

    Above normal = better

Article

The argument that broke into the weekend

A Liberal senator's warning about artists and artificial intelligence has exposed a sharp question for Australia's government: can it attract AI investment without turning the work of writers, musicians, journalists and visual artists into a resource developers can use without individual permission? Reporting published over 14–15 November 2026 described South Australian senator Andrew McLachlan's criticism of a rush to relax copyright rules for the benefit of technology companies. His intervention matters because it challenges not only Labor but also the opposition's generally welcoming approach to AI investment. The policy choice is not a simple vote for or against technology. It concerns the terms on which protected works can be used to build commercial models, the price of those permissions and the wider infrastructure required to run the systems.

What is actually being proposed

The Guardian reported that AI developer Anthropic had advanced an opt-out approach under which certain copyright material could be used for training unless rights holders objected. That is different from the traditional expectation that an organisation seeking permission to use protected work obtains it first. The government has previously rejected a broad text-and-data-mining exception, yet discussions about alternative mechanisms have continued. The Copyright Agency said in an update on 11 November that it remained unclear whether Australia would legislate changes to the Copyright Act or precisely what those changes would contain. There is therefore no enacted law granting the blanket access suggested in the debate. Confusing a company's submission with a government decision would mislead readers and creators deciding whether to negotiate licences.

The economic logic behind the disagreement

Model developers need very large quantities of text, images, audio and other material to train sophisticated systems. Licensing each work separately can create transaction costs, particularly when ownership information is hard to find or a dataset contains many contributors. AI firms argue that clearer rules could make domestic investment more attractive and reduce legal uncertainty. Creators counter that transaction costs do not nullify property rights: a valuable input remains valuable even if obtaining permission is inconvenient. The distributional consequence is central. If training rights become cheaper through law, developers may save money while some authors and publishers lose bargaining power. If licensing is too burdensome or unpredictable, investment may move to jurisdictions with other rules. Both effects are plausible; neither should be described as a quantified national outcome without evidence.

Opt-in, opt-out and collective deals are not equivalent

There are several possible architectures for consent. An opt-in system generally requires affirmative permission, giving a creator more control but creating administrative challenges for mass licensing. An opt-out system defaults toward training access unless a rightsholder takes specified steps, potentially shifting the cost of detection and objection onto individuals. A collective licensing mechanism may streamline payments by negotiating on behalf of classes of rights holders, but its fairness depends on who is represented and how revenue is distributed. McLachlan's criticism included concern that individual artists could be poorly served by a broad industry-wide compensation scheme. The government should require any proposal to explain how an unknown author, a small newspaper or a freelance photographer would discover a use, exercise rights and actually receive compensation, rather than treating the word compensation as a complete solution.

A constitutional claim is a question, not a verdict

The reporting also described claims from parts of the Coalition and the recording industry that weakening copyright protection could raise constitutional questions about the acquisition of property on just terms. Australian constitutional law does place limits on certain acquisitions of property, but it does not follow that every change in intellectual-property legislation is necessarily an unconstitutional acquisition. The legal result would depend on the design of the statute, what rights it changed, whether another party obtained a proprietary benefit and how courts applied the constitutional framework. No judicial decision on this prospective model was established by the reporting. Politicians may use constitutional language to signal seriousness; a responsible analysis must distinguish that political warning from a proven court outcome. Legal uncertainty itself can be an economic cost, especially when companies contemplate long-term datacentre investment.

Why the two major parties are exposed

Anthony Albanese (opens in a new tab) has publicly argued that creators should be compensated when their work is used. Labor also seeks investment in advanced computing and datacentres, creating a test of whether the government's eventual legal framework matches its stated commitment to creators. Opposition representatives have generally supported attracting AI development and, according to the November report, called for a durable cross-party approach. McLachlan's objections complicate the assumption that this subject can be settled by a broad agreement between party leaderships. Crossbench senators and individual backbenchers can shape legislative scrutiny even when they do not write the initial government bill. The question for voters is less about which party declares itself pro-innovation and more about who can articulate enforceable rights, credible compensation and measurable public benefits.

The creative economy's hidden participants

The debate is sometimes illustrated through famous recording artists, but many rights holders are not large corporations or household names. They include independent authors, freelance journalists, illustrators, photographers, small publishers and estates managing older works. Some have the resources to monitor training use and litigate; others do not. A formally universal opt-out right could produce sharply unequal practical protection if exercising it requires specialist knowledge or constant monitoring. Large publishers may be able to negotiate bulk deals, while an individual creator may have little leverage. On the other hand, new AI tools can help creators market their work, translate it or reduce production costs. An informed policy should examine these gains without assuming they compensate the same people who may lose licensing revenue. Distribution within the creative sector matters as much as the aggregate industry figure.

Data centres add a second set of costs

McLachlan linked copyright protection with environmental and infrastructure concerns, warning against datacentre development that harms biodiversity. That connection expands the issue beyond intellectual property. Computing infrastructure uses electricity, land, water or cooling capacity, transmission connections and sometimes public infrastructure support. The magnitude of those demands depends on location and engineering design, so it would be wrong to claim every proposed facility has the same environmental footprint. But where development consumes scarce grid capacity, the costs may be borne partly by households and other businesses if regulators allow those costs to be socialised. Where a project builds additional generation or network capacity, it may produce different outcomes. Policy should require transparent assessment of emissions, resource use, infrastructure investment and the allocation of connection costs rather than treating datacentres as either inherently harmful or inherently beneficial.

Sovereign capability versus investment incentives

Australia has strategic reasons to examine domestic compute capacity. Public agencies and businesses may want reliable access to advanced tools, local data processing and a workforce capable of evaluating systems. Yet sovereign capability does not automatically follow from the physical presence of foreign-owned servers. Relevant questions include ownership, access rights, the ability to maintain systems during international disruption, cyber resilience and how much research or skilled employment occurs locally. Ministers considering incentives should demand specific contractual or regulatory commitments in return. Otherwise a publicly supported facility might create construction activity without the lasting technological independence claimed on its behalf. The policy must be linked to energy security when additional electricity demand changes the pressure on the grid and the country's supply choices.

Lessons from other systems

The legal debate is not exclusively Australian. A legal analysis published by the International Bar Association in October described divergent approaches: American litigation about fair use, European text-and-data-mining provisions with opt-out arrangements, and continuing United Kingdom consultations. Different systems reflect trade-offs among research, investment, cultural production and property rights. Their existence does not prove that one model is universally superior or that identical statutory words will produce identical outcomes. Australian policymakers should test how any proposed exception or licensing arrangement interacts with the country's own Copyright Act 1968 and the institutions that administer rights. They should also evaluate international compatibility, because datasets, creators and model developers routinely operate across borders. A domestic statute cannot guarantee that every training activity anywhere in the world will produce a local royalty.

The calendar now matters

According to the Copyright Agency, the Joint Select Committee on Artificial Intelligence was due to present its report on 30 November. The parliamentary inquiry covers matters including competitiveness, resilience, skills and the use of Australian cultural and media content in AI training. The November reporting suggested the broader government policy package was expected before the end of 2026. These are opportunities for scrutiny, not guarantees that a complete copyright settlement will be enacted immediately. Interested parties should distinguish a consultation paper, committee recommendation, cabinet proposal, bill and final law. The steps have different legal effects. They also offer chances for authors and the public to demand clarity about the default permission rule, enforceability of opt-outs, arrangements for compensation and any exemptions for research, non-commercial use or smaller developers.

Who gains, who pays and what to watch

AI companies gain most directly from broad access to training material, clearer liability rules and cheaper access to infrastructure. Copyright owners gain from enforceable consent, licensing revenue and reliable attribution where relevant. Consumers and firms may benefit from better services, but these benefits may not be passed through in full if a market is concentrated. Artists, local electricity users and taxpayers could pay costs that are difficult to see in investment announcements. The next decisive documents will be the parliamentary committee's report and any government's draft provisions. They should be tested against practical questions: how would an independent illustrator object, how would a publisher be paid, how would compliance be audited, and who pays for extra grid capacity? Links to democratic resilience (opens in a new tab) arise when the rules governing access to news and creative work influence the financial sustainability of independent information providers. The strongest conclusion is conditional: Australia can pursue innovation and protect creators, but only if its eventual law makes the trade-offs explicit and provides rights that work in practice rather than merely in principle.

Sources

Evidence

The sources this article cites, claim-bearing material first.

  1. theguardian.com · theguardian.com/australia-news/2026/nov/15/art-over-profit-ai-policy-debate-andrew-mclachlan
  2. copyright.com.au · copyright.com.au/2026/11/ai-and-copyright-settings-update
  3. theguardian.com · The Guardian, 2 October 2026 — Anthropic opt-out proposal and public broadcasters
  4. ibanet.org · International Bar Association, 13 October 2026 — Australian AI and copyright law
Open sources page →

More analysis

Ranked by shared facts, drivers and people — never by publication date alone.