Musicians Can Verify If AI Trains on Their Tracks

Is your music being used to train AI? The Atlantic created a new search tool to help musicians check whether their recordings appear in datasets used to train AI models. The tool’s database is based on four major music datasets identified by Atlantic reporter Alex Reisner and the collections contain millions of recordings that span major artists, independent musicians and underground producers.

Those datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify, and reporting notes that AI developers use automated tools to download audio at scale; legal and ethical debates about fair use and licensing in training copyrighted material are ongoing. Some dataset sources are available for personal listening but require licensing for commercial use.

The Atlantic’s database is based on four major music datasets identified by Atlantic reporter Alex Reisner. The datasets contain millions of recordings and span major artists, independent musicians and underground producers. Two of the datasets contain around 12 million and 9 million tracks respectively, while the remaining two each include more than 100,000 songs. Artists found within the collections include Daft Punk, Aphex Twin, Radiohead, Wu-Tang Clan, and Jack Antonoff.

The datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify, making them publicly accessible in theory but not as straightforward as downloading files. Some sources, including the Free Music Archive dataset, are available for personal listening but require licensing for commercial use. According to reporting, the datasets have been downloaded thousands of times. AI developers use automated tools to download audio at scale, sometimes bypassing logins or ads, which can violate the platforms’ terms of service. Google and Stability AI acknowledged drawing on at least some material in their research papers, and legal and ethical debates around fair use have been cited by companies including Suno and Anthropic.

There are ongoing legal and ethical debates around fair use in training copyrighted material, cited by AI companies including Suno and Anthropic in court filings and public statements. Google and Stability AI acknowledged drawing on at least some material from these collections in their research papers. The datasets have been downloaded thousands of times.

Some dataset sources, including the Free Music Archive dataset, are available for personal listening but require licensing for commercial use. The datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify, which makes them publicly accessible in theory but not as straightforward to use as downloadable files. AI developers use automated tools to download audio at scale, sometimes bypassing logins or ads, which can violate the platforms’ terms of service.

The article names individual people and music groups. Named individuals include Alex Reisner and Jack Antonoff. Music groups and artists named include Daft Punk, Aphex Twin, Radiohead, and Wu-Tang Clan.

Companies and platforms mentioned include The Atlantic, Google, Stability AI, Suno, and Anthropic. Music platforms and archives cited include YouTube, Spotify, and the Free Music Archive. These organizations are referenced in relation to datasets and AI training.

Products and tools referenced include AI Watchdog and Stable Audio. The Atlantic’s new search tool for checking music in training datasets is also mentioned. No additional product names are specified in the article.

Alex Reisner said the datasets have already been downloaded “thousands of times.” The reporting presents that quoted figure as a description of dataset download activity. The article identifies Reisner as the Atlantic reporter who compiled information about the datasets. No other direct quotes from Reisner appear in the provided content.

The article includes the quoted statement “Such tools violate the terms of service of these platforms,” but the provided content does not attribute that remark to a named individual or organization. A longer quoted passage in the article addresses artistic perspectives on AI-generated work, stating that “the idea of optimising what we do is a complete miss of the entire point of what compels us in the first place,” and describing a preference for the randomness and magic of making art, but the provided content does not give an attribution for that passage.

The reporting states that AI companies Suno and Anthropic have cited legal and ethical debates around fair use in court filings and public statements. The provided content does not include direct quoted language from Suno, Anthropic, Google, or Stability AI.

Four major music datasets identified by Atlantic reporter Alex Reisner form the basis of the database used by The Atlantic’s search tool for checking whether recordings appear in AI training collections. Two of those datasets contain around 12 million and 9 million tracks respectively, while the other two each include more than 100,000 songs, giving the combined holdings a total measured in millions of recordings that span major artists, independent musicians and underground producers. The reporting notes that the datasets have been downloaded thousands of times and that The Atlantic built the search tool on that compiled database to allow musicians to check for matches.

The identified datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify rather than as packaged audio files, and some sources, including the Free Music Archive collection, are available for personal listening but require licensing for commercial use. AI developers use automated tools to download audio at scale from those linked sources, and reporting states that those automated downloads can involve bypassing logins or ads, which can violate platform terms of service. Companies including Google and Stability AI have acknowledged drawing on at least some material from the collections in their research papers, and legal and ethical debates about fair use in training copyrighted material have been cited by companies including Suno and Anthropic.

Atlantic reporter Alex Reisner identified four major music datasets that form the basis of The Atlantic’s database used in the search tool. Two of the collections contain around 12 million and 9 million tracks respectively, while the other two each include more than 100,000 songs, giving the combined holdings a total measured in millions of recordings. The collections span major artists, independent musicians and underground producers and include named artists such as Daft Punk, Aphex Twin, Radiohead, Wu-Tang Clan and Jack Antonoff.

The Atlantic’s search tool uses that database to allow musicians to check whether specific recordings appear in those datasets. Reporting notes that the datasets have been downloaded thousands of times.

The datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify rather than as packaged audio files, which means accessing the material for AI training typically involves automated processes. Some dataset sources, including the Free Music Archive collection, are available for personal listening but require licensing for commercial use.

The reporting describes AI developers using automated tools to download audio at scale, sometimes bypassing logins or ads, which can violate platform terms of service. Companies including Google and Stability AI have acknowledged drawing on at least some material from these collections in their research papers. The Atlantic’s search tool and its underlying database are presented as mechanisms for musicians to identify whether their work appears in those publicly listed datasets.

There are ongoing legal and ethical debates about the use of copyrighted music in AI training, and companies including Suno and Anthropic have cited those debates in court filings and public statements. Google and Stability AI acknowledged drawing on at least some material from the identified collections in their research papers. The reporting states that the datasets have been downloaded thousands of times.

The identified datasets are distributed as lists of links to tracks hosted on platforms such as YouTube and Spotify rather than as packaged audio files. AI developers have used automated tools to download audio at scale from those linked sources, and reporting notes that such automated downloads can involve bypassing logins or ads, which can violate platform terms of service. Some dataset sources, such as the Free Music Archive collection, are available for personal listening but require licensing for commercial use.

Named people and music artists mentioned in the article include Alex Reisner, Jack Antonoff, Daft Punk, Aphex Twin, Radiohead, and Wu-Tang Clan. These individuals and groups are referenced in reporting about the datasets identified for AI training. The article identifies Alex Reisner as the Atlantic reporter who compiled information about the datasets.

Companies and platforms named in the article include The Atlantic, Google, Stability AI, Suno, Anthropic, Free Music Archive, YouTube, and Spotify. The article references these organizations in connection with dataset sourcing, acknowledgments in research papers, and dataset distribution methods. Google and Stability AI are explicitly noted as having acknowledged drawing on at least some material from the collections in their research papers.

Products and tools listed in the article include AI Watchdog and Stable Audio. These products are referenced alongside discussion of tools and datasets used in AI music training. No additional product names are specified in the provided content.

Alex Reisner said the datasets have already been downloaded “thousands of times,” which appears as a direct quoted figure in the provided content. The provided content includes the direct quoted statement “Such tools violate the terms of service of these platforms,” and that quotation is presented without a clear attribution in the provided content.

The provided content also includes the quoted phrase “the idea of optimising what we do is a complete miss of the entire point of what compels us in the first place,” and that phrase appears without attribution in the provided content.

We (myself, the band and everyone I know) have never been looking for this work to become quicker or easier. We were never frustrated by the randomness and magic it takes. We do it for that exact reason – and without the process itself ::: nothingness. So to everyone who is gassed up about the new ways you can fake making art, by all means drive right off that cliff. We’re genuinely happy to see you go.

Alex Reisner is quoted in the provided content as saying the datasets have already been downloaded “thousands of times.” The provided content also includes the direct quoted statement “Such tools violate the terms of service of these platforms,” and that quotation appears without a clear attribution in the provided content. The provided content contains the quoted phrase “the idea of optimising what we do is a complete miss of the entire point of what compels us in the first place,” and that phrase is presented without attribution in the provided content. The provided content does not attribute those shorter quoted phrases to a named individual or organization other than the Reisner quotation.

The provided content includes a longer, unattributed quoted passage that begins “We (myself, the band and everyone I know) have never been looking for this work to become quicker or easier. We were never frustrated by the randomness and magic it takes. We do it for that exact reason – and without the process itself ::: nothingness. So to everyone who is gassed up about the new ways you can fake making art, by all means drive right off that cliff. We’re genuinely happy to see you go.” That longer passage is presented in the provided content without an attribution to a named person or organization. The provided content does not include direct quoted language attributed to Suno, Anthropic, Google, or Stability AI in the provided excerpts. The provided content therefore contains one directly attributed quotation (to Alex Reisner) and several direct quotations presented without clear attribution.