Kylie Minogue songs already used to train AI as tech firms ‘strong-arm’ Australian government, music body warns

Of the dozens of countries approached to water down their laws, Australia appears to be the only one considering it, publishers’ group says

Get our new political email, free app or daily news podcast

AI has already been trained on the works of Australian artists including Kylie Minogue and Nick Cave without compensation, the peak global body for music rights has claimed, as the Australian government has been warned against giving in to “strong-arming” from tech companies seeking to weaken copyright law.

The International Confederation of Music Publishers (ICMP) – which represents Sony, Warner and Universal as well as thousands of independent publishers on six continents – has warned that of the dozens of countries approached to water down copyright laws, Australia appears to be the only to consider doing so.

John Phelan, the ICMP’s managing director, told Guardian Australia AI companies were asking policymakers to “legitimise larceny”.

“They’re trying to strong-arm Australian policymakers into ‘if you don’t do this, we won’t invest’.”

In July, Anthony Albanese said it would be “theft” if artists did not retain control over whether their work was used to build and train AI, and what they were paid for its use.

Music, film, media and news organisations had expressed concerns the government was being lobbied by global tech companies behind closed doors to weaken Australia’s copyright laws to help train AI models, without compensating the creative sectors.

But government and opposition members of a parliamentary committee on artificial intelligence appeared to revisit the issue this week, describing the current situation as a stalemate.

Sign up for Guardian Australia’s Politics, really newsletter here

They said AI models were already training on copyrighted works in the United States without compensating creators – but that without changes to local copyright laws, those companies were withholding training frontier AI models in Australia.

In a joint letter to the prime minister this week, the inquiry’s Labor chair, Jo Briskey, and its deputy chair, Liberal senator Jonathon Duniam, recommended the government “examine current copyright settings” while ensuring creators retain control and are paid.

The independent senator David Pocock has raised concerns that the committee was coming to a position on copyright predetermined by the prime minister, after staffer edits were seen in the letter. Briskey has rejected the claim.

The ICMP has been tracking how AI companies are attempting to change copyright law in different jurisdictions to allow training. Phelan highlighted methods that the companies had used to train their models on music, such as with spreadsheets containing millions of lines of music available on YouTube.

A key claim of the AI industry is that its models are not trained on Australian content due to copyright rules. However, in evidence provided exclusively to Guardian Australia, ICMP showed Google’s Gemini was able to reproduce lyrics from Australian artists including Kylie Minogue, G-Flip and Nick Cave.

Phelan also pointed to an AI training dataset reported by the Atlantic in June, which contains more than 12m songs pulled from YouTube videos, including tracks from Minogue, Cave and Hilltop Hoods.

Phelan said 62 countries had considered making their own legal changes but none of the others had bought into AI companies’ demands.

“There isn’t a single country among the other 61 [countries] … which has introduced a new text and data mining exception for commercial purposes, or reduce the idea of exclusive rights and copyright,” Phelan said.

“The world is very much watching what Australia does.”

“It’s fair to say the global music industry, as well as all creative industries worldwide we work with, are appalled by the idea of a statutory scheme or a government deal which would interfere with how Australian industries work and contract around the world.”

Anthropic’s special envoy, Jeffrey Bleich, told the AI committee last week the company was not trying to dictate terms, but had shared information with the Australian government about training and what the implications would be “with respect to our capacity to operate here”.

“You can’t license the entire internet. Billions and billions of information exists there which is available to be viewed and studied, and that’s essentially what training is,” he said.

“Under existing Australian law, it’s our understanding that that could be determined to be illegal if we weren’t able to find a licence with every single person who had rights within Australia.”

OpenAI’s chief of economic policy, Adam Cohen, said the fair use exception in US law “fundamentally protects” the development of new technology. He pointed to agreements made with News Corp and the Guardian as examples of where the company makes deals for the use of work, despite arguing US law did not require such.

Phelan said he believed panic had set in among the AI companies over the sheer cost of investment and the need to make a return, coupled with the ongoing litigation over copyright in the US. He said they may be looking to Australia for a legislative solution.

“They’re seeing the legislative side as being the way to try to [find] a place to do business without licences, which will secure their future business model.”

In a submission to the inquiry, Guardian Australia said that undermining copyright protection would have a “profound impact on civil society” and copyright-protected material should only be available for AI training with the rights holder’s explicit prior authorisation.