The International Confederation of Music Publishers (ICMP), which represents major labels such as Sony, Warner and Universal together with thousands of independent publishers, says that artificial-intelligence systems have already been trained on the recordings of Australian musicians including Kylie Minogue and Nick Cave without any remuneration. The body warns that, among the dozens of jurisdictions approached by tech firms to relax copyright rules, Australia is the only one currently entertaining such a change.
Prime Minister Anthony Albanese has described the unauthorised use of creators’ work for AI training as theft, insisting that artists must retain control over whether their material is employed and be paid for any such use. A parliamentary committee on artificial intelligence recently revisited the issue, calling the present situation a stalemate. Members noted that AI models in the United States already train on copyrighted content without compensation, and that companies are withholding frontier AI systems from Australia until the law is altered.
Labor chair Jo Briskey and Liberal deputy chair Jonathon Duniam sent a joint letter to the prime minister urging a review of existing copyright provisions while guaranteeing that creators keep authority over their works and receive remuneration. Independent senator David Pocock raised questions about the letter’s drafting process, suggesting that edits may have been made to align with the prime minister’s position; Briskey has denied any undue influence. The correspondence reflects growing parliamentary scrutiny of tech lobbying.
ICMP’s managing director John Phelan outlined how AI firms assemble training material, citing spreadsheets that list millions of YouTube tracks and a dataset reported by The Atlantic that contains more than twelve million songs, many of which are Australian. He presented evidence that Google’s Gemini model could reproduce lyrics from artists such as Kylie Minogue, G-Flip and Nick Cave. Phelan added that none of the other sixty-one countries examined have introduced a commercial text-and-data-mining exception.
Anthropic’s special envoy Jeffrey Bleich told the committee that the firm is not dictating terms, but has briefed officials on how training operates and its impact on Australian deployment. He said licensing the whole internet is infeasible, because billions of publicly viewable items are already available for study and form the core of AI training. OpenAI’s chief of economic policy Adam Cohen invoked the U.S. fair-use principle as a key protection for innovation, citing licensing deals the company has made with News Corp and The Guardian.
Guardian Australia submitted a paper to the inquiry warning that eroding copyright protection would have a profound impact on civil society and that AI training should only proceed with the explicit prior authorisation of rights holders. Phelan warned that AI firms appear to be panicking over the cost of investment and ongoing U.S. litigation, and may be looking to Australia for a legislative fix. He quoted the companies as seeking to “legitimise larceny”, a phrase he attributed to their lobbying efforts.