{"id":4993,"date":"2026-10-19T05:13:32","date_gmt":"2026-10-19T02:13:32","guid":{"rendered":"https:\/\/mubert.com\/blog\/australia-ai-copyright-kylie-minogue-icmp"},"modified":"2026-10-19T05:13:57","modified_gmt":"2026-10-19T02:13:57","slug":"australia-ai-copyright-kylie-minogue-icmp","status":"publish","type":"post","link":"https:\/\/mubert.com\/blog\/australia-ai-copyright-kylie-minogue-icmp","title":{"rendered":"Australia AI Copyright Fight: Kylie Minogue Songs Already Used to Train AI, Publishers Say","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"\n

AI firms have already trained on Kylie Minogue’s songs, says ICMP, the global trade body for music publishers. It told The Guardian<\/a> that tech firms are “strong-arming” the government while Australia weighs its AI copyright rules. Bad timing. This post is for musicians and creators anywhere, because whatever Canberra decides could become the template the next government copies when an AI company turns up with an investment plan and a list of laws it would like changed. You get the full timeline, who said what, and what to do now.<\/p>\n\n\n\n

One of Australia’s biggest pop stars is, it turns out, also a row in a dataset. So is Nick Cave. You might be in there too.<\/p>\n\n\n\n

\"Australia\"Australia<\/figure>\n\n\n\n

Kylie Minogue songs are already in AI training data, ICMP says<\/h2>\n\n\n\n

ICMP is the International Confederation of Music Publishers. It speaks for Sony, Warner, Universal and thousands of independent publishers across six continents, according to the UA.News summary<\/a> of the Guardian piece. Its director general, John Phelan, says tech firms tie investment in Australia to new rules on copyrighted work. Translation: we’ll build here, but change the law first.<\/p>\n\n\n\n

The proof comes in two parts. First, ICMP gave The Guardian examples of Google Gemini reproducing lyrics by Kylie Minogue, G-Flip and Nick Cave. Then Phelan pointed to a training dataset The Atlantic reported on in June, which he says holds more than 12 million songs pulled from YouTube videos, with Minogue, Cave and Hilltop Hoods among them. (A training dataset is the pile a model learns from. Nobody asks the people in the pile.)<\/p>\n\n\n\n

Then there is the number ICMP clearly wanted in the headline. Phelan says 62 countries looked at changing their laws on AI training data. None of the other 61 added a new exception for commercial text and data analysis, ICMP claims. Sixty-one. So by ICMP’s count, Australia is the only one still holding the pen on AI copyright.<\/p>\n\n\n\n

What happened so far in the Australia AI copyright fight<\/h2>\n\n\n\n

This story did not start this week. Here is the timeline, oldest first.<\/p>\n\n\n\n

2025: Australia AI copyright rules get a firm no<\/h3>\n\n\n\n