
22 Sep 2026
OpenAI and Anthropic urge Australia to ease AI training copyright ban
Reuters reported that OpenAI and Anthropic, in parliamentary inquiry submissions, urged Australia to allow a narrow form of conditional approval for AI model training on creative content after the government ruled out a broad copyright exemption.
GOVERNMENT desk — the AI labs want Australian GPU barns and Australian training data; Canberra so far only offered the barns.
What Australia has already said, as Reuters reports it. A broad set of rules on AI and data centers is due to take effect next year. The government has already ruled out an exemption to copyright laws. An exemption would be permission to use protected work without the usual ask-first rule. Reuters says that ruling effectively blocks the mostly US-based companies from training models there. Training means teaching a model on existing books, songs, pictures, and other creative work.
The ask is in writing. In submissions to a parliamentary inquiry in September, on the laws that govern AI, both companies acknowledged the ban and called for it to be relaxed. A parliamentary inquiry is a committee of lawmakers collecting written views before it reports. A submission is a company’s written argument to that committee. It is not a vote.
Anthropic’s words, via Reuters. Anthropic makes the Claude chatbot. Its submission said: “Anthropic accepts that broad exceptions from copyright … have not received widespread support and have been ruled out by the government.” The alternative it offered: “we believe that the government could permit AI model training through a narrow form of conditional approval.” If Australia required “investment or other conditions designed to support the future of Australian creators and cultural endeavours,” Anthropic said it is “open to considering those conditions.” A narrow conditional approval is a limited yes, and only if the company meets conditions Canberra sets. A proposal in a submission is not an approval the government has granted.
OpenAI’s words, via Reuters. The maker of ChatGPT called for “a balanced copyright framework that allows models to learn from publicly available information” while offering rightsholders chances to collaborate. A rightsholder is the person or company that owns the copyright. “Publicly available information” is OpenAI’s phrase. Reuters does not print a definition of how wide that phrase is.
Both companies tied the copyright ask to data-center investments already underway, Reuters says. A data center here is the building full of chips that trains the models — the barn in the line above. OpenAI has signed an offtake agreement with NextDC, an Australian data-center developer, for a planned data center in Sydney. Reuters prints the ticker NXT.AX. An offtake agreement is a deal to buy a set share of a facility’s capacity. Anthropic was named last week as a partner for a data center in Queensland. A planned center is not a building this desk has seen switched on. The ticker is not a stock tip.
The clock Reuters prints. The Joint Select Committee on Artificial Intelligence is due to deliver its report in November. A joint select committee is a temporary committee drawn from both houses of Australia’s parliament. The report is still ahead. It is not this filing, and it is not a change in the copyright law.
What this card does not say. Reuters says the submissions add pressure while the government weighs AI developers against authors, publishers, and other creators. Those creators have resisted exemptions that would let their work train commercial models without consent or payment. This card does not invent that Australia changed the law, granted the narrow approval, or that creators agreed. Advocacy in a submission is not a statute.
Plain English for the rest of the card: copyright = the legal right to control a book, song, image, or other work. Exemption = permission to use that work without the usual ask-first rule. Broad exemption = a wide permission. Canberra has ruled that out. Narrow conditional approval = a limited yes, only if set conditions are met. Anthropic proposed it. Canberra has not granted it. Training = teaching a model on existing creative work. Rightsholder = the owner of the copyright. Parliamentary inquiry = a committee taking written views before it reports. Submission = a company’s written argument to that committee. Offtake = a deal to buy a set share of a data center’s capacity. Joint Select Committee on Artificial Intelligence = the parliamentary committee due to report in November. This filing is the 22 Sep Reuters account of the submissions.
RELATED
On 22 Sep 2026, Reuters reported from Sydney that OpenAI and Anthropic urged Australia to rethink a ban on using the country’s creative content to train AI models. Byron Kaye reported the story. Muralikumar Anantharaman edited it. The wire says a limited exemption could protect artists’ rights while encouraging more investment. These lines are the companies’ submissions, as Reuters reports them. Australia has not changed the law on this card.




















