Anthropic backs opt-out AI training on Australian content as ABC warns of news 'cannibalisation'

The disagreement between Anthropic and Australian media organizations over how local content is used in AI training is moving into the open. According to The Guardian, Anthropic is pushing for an opt-out model for Australian content, meaning material could be used for model training by default unless rights holders choose to withdraw.

That position runs directly against concerns raised by the Australian Broadcasting Corporation. The ABC warned that such an arrangement could cannibalise news, because outlets bear the cost of producing journalism but struggle to be rewarded when their content is used at scale for training.

The line between opt-out and opt-in consent is the crux of the dispute. For AI companies, an opt-out regime removes the need to secure permission item by item, which lowers engineering and compliance costs; for publishers, default permission shifts the burden of protecting their rights onto rights holders and creators.

Anthropic's stance draws attention because it goes straight to the boundary between Australia's content industry and the data used to train AI. Offering a concrete proposal while regulation is still unsettled suggests AI companies want a hand in shaping the final rules rather than reacting to them.

The debate is unfolding as Australia works to tighten its AI policy. With default rules for content use still undecided, the positions of AI companies and news organizations are likely to keep colliding during the legislative process, with neither side willing to concede on the default setting.

What to watch next is how Australian regulators and lawmakers respond to the split: whether they adopt the opt-out framework favored by AI companies or move toward an opt-in arrangement that is friendlier to content producers. That choice will shape where local models get their training data and will serve as a reference for other jurisdictions.