KUALA LUMPUR, July 13 — Malaysian music publishers are closely monitoring whether locally produced songs have been used without permission to train generative Artificial Intelligence (AI) models.
The concern follows an investigation by The Atlantic journalist Alex Reisner, who uncovered four massive datasets containing more than 21 million songs circulating within the AI development community.
Following the investigation, The Atlantic launched The AI Watchdog, a search engine that allows users to check whether works appear in the datasets.
A check by Malay Mail found at least 1,000 songs by Malaysian artistes, ranging from mainstream stars and veteran performers to independent acts.
A screengrab from The Atlantic’s AI Watchdog search engine showing the number of songs by selected Malaysian singers found in the datasets. National songstress Datuk Seri Siti Nurhaliza has about 106 songs listed across two datasets, including ‘Biarlah Rahsia’, ‘Percayalah’ and ‘Aku Cinta Padamu’.
The late Tan Sri P. Ramlee has about 134 songs listed in one dataset, including ‘Gelora’, ‘Mengapa Dirindu’ and ‘Getaran Jiwa’.
Malaysian R&B singer Yuna has 151 songs listed across all four datasets, while local rock band Butterfingers has 73 songs across two datasets.
Other Malaysian acts found in the datasets include Carburetor Dung, Fish Leong, Alleycats, Jess Lee, Spooky Wet Dreams, Hujan, Ziana Zain and Dayang Nurfaizah.
At least one of the datasets contains links to music hosted on widely used platforms such as YouTube and Spotify, which could explain the broad range of Malaysian discographies listed.
However, The Atlantic stressed that the presence of a song in a dataset is not definitive proof that it was used to train an AI model.
It also noted that the absence of a work from the search engine does not prove that it has never been used by an AI company.
Taja Archive Sdn Bhd Malaysia director Kimmy Khir. — Picture by Sayuti Zainudin Taja Archive Sdn Bhd Malaysia director Kimmy Khir told Malay Mail that the music publisher was aware of the AI Watchdog database and was closely monitoring developments surrounding the training datasets.
“As a music publisher, our priority is to protect the rights and interests of our songwriters and copyright owners.
“We will continue to monitor developments, understand the underlying facts, and when there is credible evidence that any works under our administration have been used without proper authorisation or in a manner that affects our rights holders, we will assess the appropriate course of action,” Kimmy said.
Taja Archive represents about 900 songwriters and administers almost 20,000 registered songs, including works by Yuna, Noh Salleh, Altimet and MimiFly.
“More broadly, we believe the industry needs greater transparency, clearer licensing practices and meaningful engagement between AI developers and rights holders so that innovation can coexist with respect for copyright,” she said.
Kimmy, who is also a council member of the Music Publishers Association of Malaysia (MPA), has been campaigning for greater understanding of AI in music and stronger regulation of the technology.
She is involved in the Malaysia Music Rights Alliance (MMRA), a task force formed by the MPA and the Recording Industry Malaysia (RIM), together with several local collective management organisations.
These include Music Authors Copyright Protection Berhad (MACP), Public Performance Malaysia Berhad (PPM) and Recording Performers Malaysia Berhad (RPM).
The MMRA recently distributed an open letter to MPs and key ministries, including the Prime Minister’s Office, the Digital Ministry and the National AI Office.
The alliance offered its expertise in developing a robust legal framework to ensure that technological innovation does not come at the expense of Malaysia’s creative economy.