Artificial intelligence has become one of the defining issues facing the music industry. Until now, however, one of the biggest questions remained almost impossible to answer:
Has my music been used to train AI?
A new tool called AI Watchdog, developed as part of The Atlantic‘s ongoing investigation into generative artificial intelligence, is beginning to provide artists with answers. By allowing musicians to search millions of songs contained within publicly identified AI training datasets, the platform offers an unprecedented glimpse into one of the industry’s most closely guarded processes.
While the tool cannot definitively prove whether a track was used to train a commercial AI model, it marks one of the first serious attempts to bring transparency to an ecosystem that has largely operated behind closed doors.
Opening AI’s Hidden Datasets
Created by researcher Alex Reisner, AI Watchdog expands upon The Atlantic‘s investigation launched in 2025 into the datasets used to build generative AI systems.
The latest version focuses specifically on music, drawing from four publicly identified datasets shared within the AI research community. Together, they contain more than 21 million tracks, ranging from internationally recognised artists to independent producers and underground electronic musicians.
The datasets originate from several different sources. Three largely reference music through links to streaming services such as YouTube and Spotify, while another draws from the Free Music Archive, a long-running collection of Creative Commons licensed recordings.
Although many of these recordings were published under licences requiring attribution or restricting commercial use, they have nevertheless become part of datasets circulating within AI development communities.
Why The Results Don’t Tell the Whole Story
The creators of AI Watchdog stress that the database should not be interpreted as definitive evidence.
A song appearing in one of the indexed datasets does not automatically mean it was ultimately used to train an AI model. Equally, if a track does not appear in the search results, that is not proof it escaped AI training.
The reality is considerably more complicated.
Most AI companies do not publicly disclose the full contents of their training datasets, often describing them as proprietary information. Researchers therefore remain dependent on datasets uncovered through academic papers, public repositories and technical documentation rather than direct disclosure from AI developers.
In other words, AI Watchdog offers visibility into part of the picture, but not yet the entire landscape.
Electronic Music Finds Itself at the Centre of the Debate
The search tool has already identified recordings by major electronic artists including Skrillex, Peggy Gou and BICEP, alongside thousands of independent producers whose music has also surfaced within the datasets.
The discovery has resonated particularly strongly within electronic music.
Unlike many other genres, electronic producers often spend years developing highly individual production techniques, synthesizer programming, drum processing and sound design. Those sonic fingerprints increasingly form the creative language that AI systems are capable of analysing and reproducing.
For many artists, the debate therefore extends beyond copyright, it touches the very identity of artistic creation.
Musicians Push Back
The publication of AI Watchdog has prompted an immediate reaction from several high-profile artists.
Producer Kenneth Blume, formerly known as Kenny Beats, publicly criticised AI music company Suno, accusing AI developers of building products at the expense of working musicians.
Meanwhile, SZA revealed that AI Watchdog had identified 238 of her songs within searchable datasets: some of which she believes were unreleased recordings.
Their responses reflect growing frustration across the creative industries as artists continue questioning how their work is collected, stored and ultimately used in commercial AI systems.
At the same time, the legal landscape remains fluid. Several major record labels have pursued legal action against AI music companies such as Suno and Udio, while others have instead entered licensing negotiations and commercial partnerships. Streaming platforms are also beginning to respond, with services like Deezer introducing AI-generated music detection technologies in an effort to improve transparency.
The Beginning of a More Transparent AI Era?
Electronic music has historically embraced technological innovation faster than almost any other genre.
From drum machines and samplers to digital audio workstations and machine-learning production tools, producers have repeatedly demonstrated that technology itself is not the enemy. The current debate instead centres on consent, transparency and attribution.
Who decides how creative work is used?
Who benefits when millions of recordings become training material?
And should artists have the ability to opt in or opt out of systems capable of generating music inspired by their own work?
AI Watchdog does not resolve those questions.
But by opening part of AI’s previously hidden infrastructure to public scrutiny, it gives artists something they have rarely had throughout this conversation:
The ability to see inside the machine.
📷 : Cover Photo Credits / generated with ChatGPT
📷 : Additional Photo Credits / generated with ChatGPT
