AI Watchdog Tracks Music Uses by AI Firms
AI Watchdog tool to track music used by AI companies helps artists check if their songs are used in AI training data and supports transparency.
The AI Watchdog tool to track music used by AI companies is an online service that enables artists to query whether their recordings appear in the datasets identified as being used to train artificial-intelligence systems. Researcher Alex Reisner created the tool on the basis of his investigation for The Atlantic, which identified four datasets spanning millions of copyrighted songs used in AI training.
The platform allows musicians to search artist names and check whether specific tracks are included in those compiled collections. The tool is presented as a resource to help artists identify potential use of their music by AI companies and to provide a searchable record of the datasets reported in the investigation.
Researcher Alex Reisner created the AI Watchdog tool following his investigation for The Atlantic that identified four datasets spanning millions of copyrighted songs used to train artificial-intelligence systems. The tool enables artists to query whether specific recordings appear within those compiled training collections.
The AI Watchdog was first launched in 2025 and has been updated to include music datasets currently in use across the AI community. The platform is presented as a searchable resource for musicians to check potential matches against the datasets reported in the investigation.
The investigation reported one dataset containing 12 million tracks, a second containing 9 million tracks, and two further collections containing over 100,000 tracks each. Reporting notes that Google and Stability have publicly used these datasets, while transparency about other companies’ usage is limited.
Google has stated it trained its own audio-generating models on materials that YouTube and Google have a right to use under their terms of service and believes this does not violate copyright laws. The AI Watchdog continues to be updated to reflect identified datasets.
The AI Watchdog tool utilizes four major datasets, each containing a vast number of copyrighted songs that have been used to train artificial-intelligence systems. The first dataset comprises 12 million tracks, making it the largest in this collection. The second dataset includes 9 million tracks, while the remaining two datasets each contain over 100,000 tracks. These collections serve as valuable resources for AI companies, allowing them to develop and enhance their audio-generating models.
Among the artists whose songs are found in these datasets are The Fall, FKA twigs, Gazelle Twin, and Objekt. These datasets have drawn significant attention, not only for their size but also for the inclusion of tracks from well-known artists. Such a large volume of music plays a crucial role in the training and development of AI systems in the music industry. However, the use of these datasets has raised concerns about copyright and transparency, as the datasets have been publicly used by companies such as Google and Stability, whereas transparency regarding other firms’ usage remains limited.
Google has publicly stated it trained its own audio-generating models on materials that YouTube and Google have a right to use under their terms of service. Google has said it believes this practice does not violate copyright laws. The company’s statement frames the sourcing of training material as covered by existing platform terms. The comment was made amid scrutiny over large music datasets used to develop AI audio models. The statement was issued in response to reporting that identified large music collections used in AI research.
Stability has publicly used the identified datasets. Reporting notes that transparency about other companies’ use of the collections is limited. The available sources do not provide detailed public statements from other companies about their use of the identified music collections. Public disclosure beyond Google’s statement is not documented in the accessible reports. The article reports that public disclosure beyond Google’s statement is not available in accessible reporting.
SZA posted: “Just checked and music AI has trained off 238 of my songs. I’m certain some unreleased. If you’re a musician and you support this degenerate shit, you’re disgusting and there’s NOTHING YOU COULD EVER SAY TO ME TO MAKE THIS OK.”
The post identifies 238 of her songs as used in AI training and explicitly references unreleased material. The language in the message expresses strong opposition to AI training practices that the artist says used her work.
No other direct artist quotes are provided in the available reporting. The available sources do not provide this information. The article places SZA’s reaction alongside reporting on dataset contents and company statements about dataset use.
CONCLUSION
The AI Watchdog tool to track music used by AI companies is available for artists to query whether their recordings appear in the datasets identified in reporting on large music collections used to train artificial-intelligence systems, enabling creators to check potential matches against those compiled collections. The article invites readers to search artists’ names on the AI Watchdog platform to determine whether specific tracks appear in those datasets. The tool, created by researcher Alex Reisner and first launched in 2025, has been updated to include music datasets currently in use within the AI community. The piece frames the AI Watchdog project in the context of transparency and advocacy for artists’ rights, and presents the platform as a searchable resource for musicians to check potential matches against the large datasets reported in the investigation.
CONCLUSION
The AI Watchdog tool to track music used by AI companies is presented here as a searchable resource that enables artists to identify whether their recordings appear in the large training datasets reported in the investigation, allowing creators to directly query compiled collections for potential matches. Researcher Alex Reisner developed the tool following his investigation for The Atlantic, and the platform was first launched in 2025 and subsequently updated that year to incorporate additional music datasets currently in use across the AI community. The article frames the AI Watchdog project in terms of transparency and advocacy for artists’ rights, describing the platform as a means for musicians and rights-holders to gain visibility into dataset contents that have been used to train audio-generating models. Readers are invited to use the AI Watchdog interface to search artist names and check whether specific tracks appear in the reported collections, with the tool positioned as a practical resource for those seeking information about potential AI training use of their work.
