Anthropic and OpenAI Urge Australia to Rethink AI Training Copyright Rules

Australia has closed off a broad copyright exemption for AI training, leaving model developers to argue for a narrower opening. In submissions to a parliamentary AI inquiry, Anthropic and OpenAI urged routes that could let models learn from certain material without restoring the approach the government rejected.

The government has ruled out an exemption that would let AI companies train on creative works without consent or payment. That moves the debate away from a blanket change to copyright law and toward terms that could limit permission or require something in return. Authors, publishers and other creators have resisted exemptions that would let commercial models use their work on those terms.

Two paths toward training access

The companies share an interest in opening a lawful path for training, but their submissions emphasize different levers. Anthropic focuses on permission tied to obligations. OpenAI centers its case on publicly available information and opportunities for rightsholders to collaborate.

Anthropic’s proposal is the more concrete trade: permission to train could be paired with investment or other support for Australian creators and cultural activities. OpenAI’s position does not use that conditional-approval language, instead pairing a training rule for public information with a role for rightsholders.

The inquiry’s next deadline

The submissions now sit with Australia’s Joint Select Committee on Artificial Intelligence, which is due to deliver its report in November. Its recommendations will indicate whether lawmakers see room for the narrower training permissions the companies want after the government rejected a broad exemption.