Australian musicians sound warning note after Nick Cave, Kylie and many more slurped into AI training tool
Paul Dempsey and Bernard Fanning are among big-name Australian musicians upset that their original songs have been found in datasets used to train artificial intelligence.
A dataset search tool recently created by US publication The Atlantic reveals millions of creative works have been scraped from the internet to train the disruptive technology.
It includes a vast catalogue of work by Australian artists, with tunes by Kylie Minogue, Powderfinger, Nick Cave and Jimmy Barnes, and novels by Thomas Keneally and Peter Carey.
Dempsey had long suspected his music was being used by AI without his permission, and says he’s found the entire catalogue of his longtime band Something For Kate as well as his solo tunes using the search tool.
“It’s frustrating this is happening. Every negotiated agreement and contract I’ve ever gone into in my career with whatever entity or record label, is all just rendered useless,” he told AAP.
“An artist’s ability to negotiate fair terms for the use of their content is just being ripped away from them.”
Using original songs to produce robotic AI content is ultimately dehumanising, Bernard Fanning argued.
“Do we want robots telling our stories and synthesising our feelings? Because it’s not human. The whole point of art is to humanise our feelings, to express how we’re feeling across the whole range of emotions,” he told AAP.
“Robots aren’t alive, they don’t experience, they just aggregate - and the idea of that sucks.”
Songwriter Darren Hayes found in the datasets the entire output of his 30-year recording career, including Savage Garden hits such as Truly Madly Deeply, and recently took to Instagram to express his fury.
“I absolutely feel violated that all of the hundreds and hundreds and hundreds of hours, blood sweat and tears that I’ve put into my music along with every other musician, has been stolen and served up like French fries to a piece of software that spits out shit,” he said.
The Australian songs are contained in two datasets, the first assembled by a group of researchers known as Sleeping AI.
Sleeping-DISCO-9M comprises 9.7 million music tracks from YouTube plus lyrics from Genius.com, while a second dataset LAION-DISCO-12M was created by Germany-based group LAION, using 12.3 million YouTube tracks.
The Atlantic cautioned that AI companies might omit works when training their models, so the inclusion of songs in datasets is not definitive proof they have been used.
The datasets are proof of the theft of creative work, according to music licensing organisation APRA AMCOS, which represents 128,000 members in Australasia.
“Major tech platforms have not come to the table. Not once. Instead they have lobbied governments, circulated policy papers, and proposed solutions designed to extinguish any obligation to pay,” said chief executive Dean Ormston.
Australia’s intellectual property laws hold that permission should be granted and terms such as payment agreed on before copyright works are used, but the IT industry has pushed for text and data mining exemptions to the laws.
In August 2025, the Productivity Commission floated changes which would have legalised AI companies using content without paying creators, but the federal government ruled out the changes in October.
Dempsey is midway through his Shotgun Karaoke regional tour of Australia, and says genuine artistic expression comes from human experience, not artificial intelligence.
“We can trigger huge emotional responses in each other through art, and I don’t know that that’s going anywhere, it’s just going to be flooded with all this other shit,” he said.
Read the full story at The Guardian ↗ · The Guardian ↗ · The Guardian ↗ · The Guardian ↗
Australian musicians have identified their recordings in publicly accessible datasets used to train artificial intelligence systems. Two large datasets—one containing 9.7 million tracks and another with 12.3 million—were assembled by research groups using content from YouTube and lyrics databases. Artists including Paul Dempsey, Bernard Fanning, and Darren Hayes say they were not consulted or compensated. Under Australian law, copyright holders must grant permission and agree on terms before their work is used commercially. The tech industry has sought exemptions to these requirements through policy proposals and government lobbying. In October 2025, the federal government declined to adopt changes that would have permitted AI companies to use copyrighted material without creator consent or payment. Music licensing organisations and affected artists contend this represents unauthorised use of creative work.
Read the full story at The Guardian ↗ · The Guardian ↗ · The Guardian ↗ · The Guardian ↗
Paul Dempsey and Bernard Fanning are among big-name Australian musicians upset that their original songs have been found in datasets used to train artificial intelligence.
A dataset search tool recently created by US publication The Atlantic reveals millions of creative works have been scraped from the internet to train the disruptive technology.
It includes a vast catalogue of work by Australian artists, with tunes by Kylie Minogue, Powderfinger, Nick Cave and Jimmy Barnes, and novels by Thomas Keneally and Peter Carey.
Dempsey had long suspected his music was being used by AI without his permission, and says he’s found the entire catalogue of his longtime band Something For Kate as well as his solo tunes using the search tool.
“It’s frustrating this is happening. Every negotiated agreement and contract I’ve ever gone into in my career with whatever entity or record label, is all just rendered useless,” he told AAP.
“An artist’s ability to negotiate fair terms for the use of their content is just being ripped away from them.”
Using original songs to produce robotic AI content is ultimately dehumanising, Bernard Fanning argued.
“Do we want robots telling our stories and synthesising our feelings? Because it’s not human. The whole point of art is to humanise our feelings, to express how we’re feeling across the whole range of emotions,” he told AAP.
“Robots aren’t alive, they don’t experience, they just aggregate - and the idea of that sucks.”
Songwriter Darren Hayes found in the datasets the entire output of his 30-year recording career, including Savage Garden hits such as Truly Madly Deeply, and recently took to Instagram to express his fury.
“I absolutely feel violated that all of the hundreds and hundreds and hundreds of hours, blood sweat and tears that I’ve put into my music along with every other musician, has been stolen and served up like French fries to a piece of software that spits out shit,” he said.
The Australian songs are contained in two datasets, the first assembled by a group of researchers known as Sleeping AI.
Sleeping-DISCO-9M comprises 9.7 million music tracks from YouTube plus lyrics from Genius.com, while a second dataset LAION-DISCO-12M was created by Germany-based group LAION, using 12.3 million YouTube tracks.
The Atlantic cautioned that AI companies might omit works when training their models, so the inclusion of songs in datasets is not definitive proof they have been used.
The datasets are proof of the theft of creative work, according to music licensing organisation APRA AMCOS, which represents 128,000 members in Australasia.
“Major tech platforms have not come to the table. Not once. Instead they have lobbied governments, circulated policy papers, and proposed solutions designed to extinguish any obligation to pay,” said chief executive Dean Ormston.
Australia’s intellectual property laws hold that permission should be granted and terms such as payment agreed on before copyright works are used, but the IT industry has pushed for text and data mining exemptions to the laws.
In August 2025, the Productivity Commission floated changes which would have legalised AI companies using content without paying creators, but the federal government ruled out the changes in October.
Dempsey is midway through his Shotgun Karaoke regional tour of Australia, and says genuine artistic expression comes from human experience, not artificial intelligence.
“We can trigger huge emotional responses in each other through art, and I don’t know that that’s going anywhere, it’s just going to be flooded with all this other shit,” he said.
Read the full story at The Guardian ↗ · The Guardian ↗ · The Guardian ↗ · The Guardian ↗
Paul Dempsey, Bernard Fanning, Nick Cave, Kylie Minogue, Darren Hayes, and other Australian musicians have found their work in AI training datasets Two datasets—Sleeping-DISCO-9M (9.7 million YouTube tracks plus Genius.com lyrics) and LAION-DISCO-12M (12.3 million YouTube tracks)—contain Australian music The Atlantic's dataset search tool allows public identification of which creative works were scraped for AI training Australian copyright law requires permission and negotiated payment terms before copyrighted works are used Tech industry groups have lobbied for text and data mining exemptions to copyright law The Australian government rejected Productivity Commission proposals in October 2025 that would have legalised unpaid AI training on copyrighted content APRA AMCOS represents 128,000 music creators and characterises inclusion in datasets as theft This practice renders artists' negotiated contracts useless AI-generated content using human artistic work is dehumanising because robots aggregate rather than experience The inclusion of songs in datasets constitutes proof they have been used in AI models
Read the full story at The Guardian ↗ · The Guardian ↗ · The Guardian ↗ · The Guardian ↗
- Multiple Australian musicians including Nick Cave, Kylie Minogue, and Paul Dempsey have discovered their work in AI training datasets without permission
- Two datasets (Sleeping-DISCO-9M and LAION-DISCO-12M) containing millions of music tracks scraped from YouTube and lyrics sites are being used to train AI systems
- Australian copyright law requires permission and negotiated terms before creative works are used, but tech companies have lobbied for exemptions
- The Australian government rejected proposed changes in October 2025 that would have legalised AI training on copyrighted content without creator compensation
- Artists argue the practice undermines their ability to negotiate fair terms and reduces human artistic expression to algorithmic aggregation