Leadership1 publisher3 min readPublished
Anthropic asks Australia to let AI train on copyrighted work unless owners opt out
Anthropic told an Australian inquiry a broad copyright exemption is off the table and proposed opt-out 'conditional approval' for AI training instead. The ABC and SBS want the onus on AI firms, so the inquiry is now deciding who pays to police training bots.
The Board Room · Leadership desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Anthropic proposed robots.txt "do not crawl" signals as the way rightsholders would opt out, and said it was open to supporting Australian creators.
- The ABC told the federal parliamentary inquiry that AI companies should face the same copyright, defamation and privacy rules as media outlets.
- The ABC and SBS asked for AI firms to be added to the news bargaining incentive, which would require commercial deals with large media outlets to support journalism.
- The joint select committee on AI holds hearings in Sydney next week, with executives from Anthropic and OpenAI due to attend.
Compiled by The Board RoomSomething wrong?How this is made
Why it matters
- cost An opt-out would leave Australian publishers paying on an ongoing basis to find and block each new training crawler, a cost SBS says belongs with the AI companies.
- decision AI developers weighing Australian datacentre spending now have to price two outcomes: training subject to opt-outs, or required commercial deals with large news outlets.
- precedent Adding AI firms to the bargaining incentive would put them under the regime already forced on Facebook, Google, TikTok and Snapchat, widening a platform rule to cover model makers.
Anthropic's concession leaves the Australian argument with one variable: where the default sits. "Anthropic accepts that broad exceptions from copyright, such as a text and data mining exception, have not received widespread support and have been ruled out by the Government. As an alternative, we believe that the Government could permit AI model training through a narrow form of conditional approval," the company submitted [2]. The Labor government had already ruled out that exemption, with Anthony Albanese promising "the strongest possible protection" for Australian creatives [3]. The question left open was an opt-out, letting copyright holders choose whether AI companies can access their work [15].
The trade-off between the two camps is who has to act and who pays for it. An opt-out makes the rightsholder move to stop training. A bargaining obligation makes the AI company move to get a deal [4] [9].
SBS made the sharpest case against the opt-out. "Robots.txt mechanisms are routinely ignored or bypassed, and the blocking of ever-evolving training bots increases the burden on rightsholders to monitor and update protections, when the onus should be on AI companies to avoid wrongdoing," the broadcaster said in its submission [5]. I think SBS is right on enforcement. A crawler signal binds only the crawlers that choose to honour it, and a publisher finds the rest by watching its own traffic. Anthropic's reply so far is a statement of intent: it wants Australia to "provide legal certainty under copyright that brings AI developers to the table while advancing its national cultural interests" [14].
The ABC's case reaches past training data to what AI services put out. If the inquiry accepts that framing, the question moves from what goes into a model to who answers for what comes out of it. The broadcaster called the current system "anti-competitive" [7]. "As publishers, AI-based services are currently unregulated and receive more favourable treatment than established Australian media. This unequal treatment will accelerate the cannibalisation of the Australian news industry," the ABC said in its submission [8].
Anthropic paired its copyright request with an investment pitch. It talked up plans to invest billions in Australia, including large new datacentres, and urged the government to support "frontier model training" [12]. In my view those datacentre plans are the company's main leverage with a government that has already closed off the broad exemption [3]. The same submission described "light and shade" in AI's development and warned of "serious implications for Australia's national security, the resilience of its critical infrastructure and the safety of its people" [13].
The inquiry is gathering positions this quarter, and the reporting does not give a date for a government decision. Whichever default it picks becomes the starting point for the licensing talks that follow. A publisher under an opt-out negotiates with the threat to block, and keeps paying to make that threat credible. An AI developer under the bargaining incentive negotiates already owing a deal [9].
What to watch
- Whether Anthropic's executives spell out the conditions behind 'conditional approval' when they give evidence in Sydney.
- Whether the government signals that the news bargaining incentive will cover AI developers as well as platforms.
- Whether OpenAI, also appearing before the committee, backs Anthropic's opt-out proposal or argues for a different model.