AI giant wants copyright certainty, even as litigation soars over AI training usage and politician calls proposal a “dirty deal”
Anthropic has told the federal government that its proposed US$15bn ($21.6bn) investment in Australian AI infrastructure hinges on the government providing greater certainty around copyright law, according to a Treasury briefing note prepared for Treasurer Jim Chalmers ahead of a meeting with Anthropic chief executive Dario Amodei. The note, released under freedom of information laws and reported by the Australian Financial Review, states that Anthropic's investment in "AI model development capability and associated infrastructure, like data centres, is contingent on clarity of copyright settings," driven by the company's desire for "certainty over their liability to rights holders."
Notably, the briefing indicates Anthropic stopped short of seeking an outright exemption from copyright law — an option the Albanese government has already ruled out — and instead sought clarity on its existing obligations, including how it might deal with what the company describes as a "long tail" of smaller rights holders whose works are difficult to identify and licence individually. Treasury reportedly advised Chalmers to direct Amodei to the Attorney-General's Department, which is running the government's AI and copyright reform process.
The disclosure lands in the middle of an active, and increasingly contested, law reform process. Attorney-General Michelle Rowland ruled out a text and data mining (TDM) exception in October 2025, after the option was floated by the Productivity Commission and drew fierce opposition from creative and media industries. Since then, Rowland's department has been consulting on alternative models for compensating rights holders, reportedly canvassing three broad options: a statutory licensing scheme with set payments; a collective licensing model run through bodies such as the Copyright Agency; and the existing voluntary, negotiated licensing framework. The Treasury briefing also indicates the government is considering a new enforcement regime to make it easier for smaller rights holders to pursue technology companies for uncompensated use of their work. Separately, legal commentators tracking the reform process — including Bird & Bird — have suggested a small claims forum is among the mechanisms the Copyright and AI Reference Group could examine, though this has not been confirmed as government policy. I do not have a verified source confirming Rowland has formally committed to this specific mechanism, and readers should treat it as an unresolved possibility rather than settled reform.
The Treasury briefing pushes back directly on one of the arguments frontier AI developers have used to justify reform, noting that Anthropic's characterisation of AI training as "highly transformative" and covered by the US fair use doctrine is legally unsettled. The note points to the volume of active litigation — an estimated 81 AI-copyright lawsuits before US courts as of February 2026 — including Anthropic's own case, which concluded last year with the company agreeing to pay $1.5bn to settle a class action brought by authors over the use of pirated book copies in model training, equivalent to roughly $3,000 per book across 500,000 titles. (The AFR briefing note does not specify the currency for this figure; based on public reporting of the underlying US case, this is very likely US dollars, but that has not been independently confirmed for this article.) The presiding US court found that training AI systems on lawfully acquired copyrighted books did not itself breach copyright, but that using pirated copies did.
The stakes have been sharpened by a separate allegation, aired by independent senator David Pocock, that a cabinet-level proposal under consideration would offer a copyright carve-out in exchange for at least $50bn in data centre investment and roughly $350m a year for a creative-sector fund — a proposal Pocock has labelled "the ultimate dirty deal." The government and named ministers have rejected the characterisation as inaccurate, and industry groups have previously mobilised against similar suggestions: eighteen media and creative organisations, including News Corp, Nine Entertainment and the Australian Writers' Guild, issued a joint statement in May opposing any reopening of copyright settings, arguing that licensing deals with AI developers are already proving workable across news, music and publishing.
Cabinet is reportedly split on the path forward, with Industry Minister Tim Ayres and Assistant Minister for the Digital Economy Andrew Charlton seen as more receptive to the investment case, against a more protective posture from Rowland and Arts Minister Tony Burke. Albanese is expected to address the tension — without announcing specific copyright changes — in a speech this week billed as a broader statement of the government's AI policy direction, including its approach to data centre approvals and "social licence" expectations for developers.
For copyright, media and technology lawyers, the immediate significance lies less in any single company's negotiating position than in the direction of travel for the Copyright Act 1968 (Cth). None of the reform options on the table — statutory licensing, collective licensing, or an enhanced voluntary regime with better enforcement — would replicate the fair use or opt-out text and data mining exceptions available to AI developers in the United States or European Union, meaning Australian-based or Australia-facing AI training is likely to remain subject to a permission-based regime for the foreseeable future. Practitioners advising rights holders, AI developers or data centre investors should expect continued uncertainty on timing, with the Attorney-General's Department yet to commit to a legislative timetable, and should watch closely for further detail following Albanese's address and any formal response from the Copyright and AI Reference Group.
