SH Labs, SubmitHub’s AI Music Detector, says it analysed more than one million music releases during July and found that nearly 40% showed signs of AI involvement. Its reported classifications split the total into 23.2% labelled fully AI-generated and 15.3% labelled as AI-generated audio subsequently modified or processed by humans.
Those figures should not be read as proof that 40% of all music released in July was made by AI. They describe the output of one detection system across the releases it analysed, using classifications and methodology that are not fully detailed in the supplied reporting. Still, the scale of the sample makes the data relevant to independent producers, distributors and curators dealing with AI disclosure and automated content checks.
Key Takeaways
- SH Labs says its detector processed more than one million releases in July.
- It classified 23.2% of analysed recordings as fully AI-generated and 15.3% as modified or hybrid AI-generated audio.
- The combined 38.5% figure is a detector result, not independently established proof of how each track was created.
- SH Labs says 31% of artists whose music it identified as containing AI-generated audio declared that they had not used AI during submission.
- Producers using generative audio should keep clear records of their process and complete platform disclosures accurately.
What SH Labs is claiming
According to the reported SH Labs data, the detector is designed to separate recordings it considers fully AI-generated from tracks containing AI-generated material that has been edited, processed or combined with human-made elements.
That distinction matters more than a simple AI-or-not label. A fully prompt-generated song and a conventional production containing one generated vocal texture, stem or instrumental element present very different creative, contractual and rights questions. A detector may be useful to platforms wanting to apply different policies to those cases, but the classification itself does not establish authorship, ownership or copyright status.
SH Labs also says the technology is used by organisations including Bandcamp, GEMA’s MusicHub, Traxsource, distributors, digital platforms and one major label. Producers can view the company’s stated service and positioning at SH Labs.
Detection signals are not the same as evidence
Automated AI detection is likely to become part of more music-submission workflows, particularly where platforms and curators want to filter large volumes of uploads. But an AI flag should be treated as a signal for review, not a final finding about how a recording was made.
The supplied report says SH Labs achieved more than 99% accuracy on a confidential industry benchmark. It does not provide the benchmark dataset, testing conditions, error rates by genre, or the threshold used to distinguish generated audio from heavily processed human performances and productions. Without those details, producers and rights holders cannot independently assess how the detector performs on their type of work.
That limitation is especially relevant in electronic music, where aggressive vocal processing, resampling, synthesis, spectral manipulation, stem separation and restoration tools can produce audio far removed from an untreated recording. None of those production methods alone demonstrate generative-AI use.
The reported 31% mismatch between artists’ declarations and detector findings is therefore notable, but it does not on its own establish deliberate non-disclosure. It could reflect differing definitions of AI use, unclear submission questions, hybrid workflows, or incorrect detector classifications. The report does not break that figure down further.
What independent producers should keep on file
For producers releasing music that includes generative or AI-assisted elements, good documentation is becoming a practical safeguard. Retain DAW sessions, bounced stems, MIDI, version histories, sample licences and notes on which tools were used. This will not necessarily prevent an automated flag, but it gives you a clearer record if a distributor, label, collaborator or curator asks how a track was made.
Be precise when a submission form asks about AI. Check whether it refers specifically to generated audio, vocals, composition, artwork, mastering assistance or any machine-learning-based tool. The distinction matters: using an AI-enabled repair, separation or mastering feature is not necessarily equivalent to uploading a fully generated recording, but a platform’s own policy may define the category differently.
Producers working with collaborators should also put AI use into their project paperwork. Agree who created any generated elements, what service was used, whether its terms permit commercial release, and whether the contributor has disclosed that use to the rest of the team. That is sensible release administration even before a distributor or rights organisation asks questions.
What to watch next
The useful development here is not a definitive industry-wide percentage; it is the growing likelihood that release platforms and music businesses will use automated signals alongside artist declarations. SH Labs’ data suggests that hybrid productions are a central part of that challenge, rather than a marginal edge case.
For now, the practical approach is straightforward: disclose accurately where required, keep production records, and do not assume a detector label settles the creative or legal history of a song.