Anthropic Asks Australia for Opt-Out Rule on AI Training
Anthropic has asked the Albanese government to consider conditional approval for technology companies to train AI models on Australian copyrighted works under an opt-out model. The proposal would allow training by default, with a way for rights holders to exclude their material, according to The Guardian. Opt-out means permission is assumed unless the owner objects, like staying on a list unless you unsubscribe.
Anthropic accepted that a broader copyright change, called a text and data mining exception, has been ruled out. That exception would have created a general right to copy large collections of text and images for computer analysis. The company said it wants legal certainty under copyright that brings AI developers to the table while supporting Australia's cultural interests. The Labor government has ruled out a text and data mining exemption for AI training.
Anthropic said it plans to invest billions in Australia, including large new data centres, and it urged support for frontier model training. Frontier models are the most capable AI systems, built with very large data sets and computing power. Canberra has not agreed to the opt-out framework.
The ABC, in a submission to a federal parliamentary inquiry, said AI companies should follow the same copyright, defamation and privacy rules as media outlets. Defamation is the law on false statements that harm reputation. The broadcaster warned that different treatment for unregulated AI services will speed up the cannibalisation of the Australian news industry. It argued AI systems should carry the same liability as publishers, not be treated as neutral intermediaries.
The ABC and SBS suggested the government include AI companies in the news bargaining incentive. That scheme requires large platforms to make commercial deals with media outlets to support journalism. SBS said AI large language models, the systems behind tools such as ChatGPT or Claude that generate answers from patterns in text, harm the sustainability of Australian news by using news content without permission, attribution or compensation.
Australia's joint select committee on AI will hold hearings in Sydney next week, with executives from Anthropic and OpenAI expected to attend. Separately, Anthropic will not appear before an Australian Senate committee hearing on AI on October 1, 2026, according to Reuters. The two events involve different committees and dates.
Anthropic denied it has any deal or negotiations with the ABC to license its archive of content, according to AFR. The company said no licensing arrangement is in place.
The broader context here is a negotiation over who sets the default. An opt-out regime keeps the burden on publishers and creators to monitor and object, while preserving large-scale ingestion for model developers. A permission-first regime reverses that burden and requires licensing before ingestion. Both sides invoke sustainability, one for frontier capacity and data-centre investment, the other for news production and cultural production.
Looking at what this means for policymakers, the immediate test is whether legal certainty can be constructed without reopening the text and data mining fight. Conditional approval suggests conditions on transparency, remuneration or access for Australian researchers and cultural institutions could be attached. The public broadcasters are pushing to tie any certainty for developers to enforceable obligations on compensation, attribution and parity under media law. The Sydney hearings will show how far either side moves from those opening positions.
