Anthropic seeks an opt-out path to train on Australian works

Broadcasters want equal rules

ABC told a federal parliamentary inquiry that AI companies should be subject to the same copyright, defamation and privacy laws as media organisations. It said the regulatory gap could accelerate the decline of Australia’s news industry.

ABC and SBS also want AI companies included in a program that encourages commercial deals between large technology platforms and major news publishers to support journalism. ABC says AI services are not regulated as publishers and receive more favourable treatment than traditional Australian media.

SBS warned that large language models such as ChatGPT and Claude threaten the sustainability of Australian news outlets. It also said using news without permission, attribution or compensation could weaken the reliability of information and public trust.

The hearing comes as Anthropic seeks a route to training on Australian copyrighted material. Australia’s Labor government has rejected a text-and-data-mining exception, and Prime Minister Anthony Albanese has promised authors “the strongest possible protection.” Debate continues over other copyright changes, including whether rights holders should be able to choose whether AI companies use their work.

Anthropic’s narrower proposal

Anthropic says AI could transform Australia’s economy, raise productivity and help address major national problems. It also warned of serious risks to national security, critical infrastructure and public safety. The company says it plans to invest billions of dollars in Australia, including in large data centres, and urged the government to support training advanced models.

Rather than a broad copyright exception, which Anthropic acknowledged lacks wide support and is no longer under consideration by the government, it proposed a narrower, conditional permission for model training. One possible mechanism would use robots.txt signals to tell bots which material they may access. Anthropic said it is willing to support Australian creators.

SBS doubts an opt-out can work in practice. It says robots.txt rules are often ignored or bypassed, while rights holders would have to keep monitoring changing training bots and updating protections. In SBS’s view, AI companies—not creators—should be responsible for preventing violations.

The disagreement is partly about who carries the burden. An opt-out gives rights holders a way to say no, but it also asks them to keep that refusal effective as bots and practices change. I think that is the unresolved weakness in Anthropic’s proposal: legal certainty for model developers may come at the cost of ongoing work for the people whose material they want to use.

The bargaining question

ABC called the current system anti-competitive and, alongside SBS, renewed its call to bring AI companies into the news-deals program. Such agreements are already required of large technology platforms, including Facebook, Instagram, Google, TikTok and Snapchat.

The announcement leaves a harder question open: what would count as fair compensation when a model trains on news, and who would be responsible for securing it? The broadcasters’ answer is to extend the existing bargaining framework to AI companies. Anthropic’s proposal instead centres on permission signals and a conditional legal route. Those are different answers to the same problem—and the Sydney hearing will put both in front of lawmakers.