The debate over how Australia should regulate artificial intelligence has intensified after two of the world’s leading AI companies, OpenAI and Anthropic, urged the government to reconsider restrictions that prevent them from using Australian creative works to train their AI models.

In separate submissions to a parliamentary inquiry, the companies acknowledged the government’s decision not to introduce a broad copyright exemption for AI training but proposed a narrower arrangement that could allow model developers to access copyrighted material under specific conditions.

The proposals come as governments globally grapple with how to regulate rapidly advancing AI technologies while attracting the investment associated with the industry.

Australia has announced plans for a broad framework covering artificial intelligence and data centres, with the new rules expected to take effect from next year. However, the government has already ruled out an exemption to existing copyright protections that would allow AI companies to freely use creative works for training.

The decision has become a major point of contention between technology companies and Australia’s creative industries, with authors, publishers and other rights holders opposing arrangements that could allow commercial AI developers to use their work without permission or compensation.

Anthropic proposes conditional access

In its submission to the Joint Select Committee on Artificial Intelligence in September, Anthropic, the company behind the AI chatbot Claude, acknowledged the government’s position but proposed what it described as a narrower alternative.

“Anthropic accepts that broad exceptions from copyright ... have not received widespread support and have been ruled out by the government,” the company said.

“As an alternative, we believe that the government could permit AI model training through a narrow form of conditional approval,” it added.

Anthropic suggested that access could be tied to requirements intended to benefit Australia’s creative sector.

“If Australia required ‘investment or other conditions designed to support the future of Australian creators and cultural endeavours ... we are open to considering those conditions,’” the company said.

The proposal would potentially give the government a way to maintain copyright protections while allowing AI developers to train their models under a more tightly controlled system.

OpenAI calls for ‘balanced’ copyright framework

OpenAI, the creator of ChatGPT, also called for changes to Australia’s approach, arguing that AI models should be able to learn from information that is publicly available while giving copyright holders opportunities to participate.

The company called for “a balanced copyright framework that allows models to learn from publicly available information” while offering rightsholders opportunities to collaborate.

The submissions place the two companies among technology firms seeking greater access to data as governments consider how existing copyright laws should apply to AI systems.

At the heart of the dispute is the question of whether AI companies should be permitted to use copyrighted books, articles, images and other creative works to train commercial models without obtaining individual permission from rights holders.

Creators and publishers in Australia and elsewhere have increasingly challenged such practices, arguing that commercial AI systems should not be trained on their work without appropriate consent or payment.

AI investment adds pressure

The copyright debate is also unfolding alongside significant investments in Australia’s AI and data-centre infrastructure.

OpenAI has signed an offtake agreement with Australian data-centre developer NextDC for a planned data centre in Sydney, while Anthropic was named last week as a partner for a proposed data-centre project in Queensland.

The infrastructure projects highlight the competing interests facing Australian policymakers as they attempt to develop rules for an industry expected to attract substantial investment while protecting local businesses, workers and creative industries.

Governments around the world are similarly trying to balance regulation with the economic opportunities presented by artificial intelligence. Economists expect the global AI industry to generate potentially hundreds of billions of dollars in economic value by the end of the decade.

For Australia, the challenge is particularly focused on how to encourage investment without weakening protections for authors, publishers, artists and other creators.

The latest submissions are therefore likely to add to pressure on the government as it considers how its proposed AI framework should address copyright and model training.

The Joint Select Committee on Artificial Intelligence is expected to submit its report in November, when its recommendations could provide further direction on how Australia intends to reconcile the interests of AI developers with those of copyright holders and the wider creative economy.