Anthropic urges Australia to allow opt-out AI training

Artificial intelligence firm Anthropic urged the Australian government to grant conditional approval for tech companies to train models on copyrighted works under an opt-out model. The proposal, made in a submission to a federal parliamentary inquiry, followed the Australian government ruling out broad text and data mining exceptions.

Anthropic stated that AI could transform the Australian economy and drive productivity, while noting plans to invest billions in the country, including large new data centres. As an alternative to broad exemptions, the company proposed using robots.txt signals to let copyright holders opt out, while seeking legal certainty for frontier model training.

Australian public broadcasters ABC and SBS opposed the proposal in their own submissions. The ABC warned of the cannibalisation of mainstream news and called for AI companies to face the same copyright, defamation, and privacy rules as publishers. SBS said robots.txt signals are routinely bypassed and create an unfair burden for creators.

Both broadcasters urged the government to include AI companies in the news bargaining code, requiring them to negotiate commercial compensation deals with news outlets. Australia's joint select committee on AI will hold hearings in Sydney, with executives from Anthropic and OpenAI scheduled to attend.