Anthropic has submitted evidence to Australia's Joint Select Committee on Artificial Intelligence, proposing that the government adopt a "conditional licence" for AI model training. Under the plan, tech companies could train models on copyrighted Australian content by default, and rights holders who object could block AI crawlers using a robots.txt file — the mechanism commonly known as "opt-out."
According to The Guardian, Anthropic acknowledged in its submission that it isn't seeking a blanket copyright exemption. The Australian government already ruled out a text-and-data-mining (TDM) exception last year, meaning AI companies won't be allowed to train on copyrighted works without permission and without conditions. Anthropic's proposal is a narrower version of that idea, and the company said it is willing to support Australian creators.
Two public broadcasters push back
The Australian Broadcasting Corporation (ABC) and the Special Broadcasting Service (SBS) both opposed this approach in their own submissions.
The ABC argues the current arrangement is unfair to journalism and anti-competitive, and it wants AI companies brought into Australia's News Bargaining Incentive, the mechanism that currently covers Facebook, Google and TikTok and is meant to push large platforms to pay news organisations. The ABC also wants AI companies held to the same copyright, defamation and privacy rules that apply to media outlets.
"This unequal treatment will accelerate the cannibalisation of the Australian news industry."
SBS's submission names ChatGPT and Claude specifically, saying large language models use news content without permission, attribution or payment, which it says threatens the industry's sustainability and public trust. On robots.txt, SBS is blunt: the mechanism is frequently ignored or bypassed, and the responsibility for avoiding infringement should sit with AI companies rather than requiring publishers to police the entire open web.
Where opt-out rules stand elsewhere
Opt-out style rules already exist in other jurisdictions. Article 4 of the EU's 2019 Copyright Directive permits text and data mining for commercial purposes unless a rights holder has reserved their rights in a machine-readable way, and the EU AI Act requires providers of general-purpose models to comply with those reservations. The UK government consulted on a similar scheme in late 2024, which drew public opposition from large numbers of musicians, writers and publishers and has not been implemented.
Japan has gone further: Article 30-4 of its Copyright Act gives a fairly broad exception for uses aimed at information analysis. The United States has no statutory training exception at all, and disputes are fought case by case in court under fair use. Anthropic itself has already paid a price in the US — it agreed last year to a $1.5 billion settlement with a class of authors after training data was found to include books downloaded from pirate sites.
Against that backdrop, Australia's current position is distinctive: the government has already ruled out a TDM exception, the opt-out mechanism is still under discussion, and the "pay first" model favoured by the ABC and SBS has an existing framework — the News Bargaining Incentive — it could slot into. The government has not given a timetable for which of the three paths will make it into a final policy.
A hearing next week
The committee will hold a hearing in Sydney next week, and both Anthropic and OpenAI are expected to send executives. Separately, it has been reported that OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei will not attend a separate Senate inquiry. What the two companies will be asked at the Sydney hearing, and whether either will budge on the payment demands, is not yet public.
Sources: The Guardian, The Next Web, CocoLoop, public submissions to Australia's Joint Select Committee on Artificial Intelligence; ABC and SBS submissions for quotes, EU Copyright Directive Article 4 and Japan's Copyright Act Article 30-4 for statutory details.