Music & AI
AI detectors flag a handmade album and 1982 recordings as synthetic

A test by the agency International Artist Marketing, reported by Digital Music News, shows SubmitHub and other tools attributing to AI synth tracks made without automatic generation, including recordings more than forty years old.
Twenty-three tracks out of twenty-three classified by SubmitHub as potentially AI-generated, and seven original 1980s recordings flagged as "possible AI/hybrid". That is the result of a check carried out by the agency International Artist Marketing (IAM) and reported on 4 November 2026 by Digital Music News, which reopens a concrete issue for anyone producing electronic music: tools that promise to recognise artificially generated audio can get it wrong, and the error falls on independent musicians.
The Coloured Tears case
The album at the centre of the test is DECAY & TIME, a double tribute to Twice a Man, the Swedish post-punk group of the 1980s, released on 26 June 2026 by Hamburg musician André Grafert under the project name Coloured Tears: according to its Bandcamp page it collects 23 reinterpretations of the band's songs released between 1982 and 1988. Grafert says he built it note by note in Cubase, with Sequential synths (Prophet X, Prophet 12, Pro 3, Rev2), Native Instruments Kontakt libraries and his own voice, without generative AI. IAM analyst Ines Weidemann ran all the tracks through SubmitHub and spot-checked other detectors; according to DMN, these too judged them to be predominantly AI. The tools cited in the article are SubmitHub, Let's Submit, TrackVerifier and ACRCloud.
The most telling detail concerns the cross-check: Weidemann also analysed the original Twice a Man recordings, i.e. the versions the covers are based on. According to DMN, SubmitHub marked seven of them as possible hybrids, and two were even linked to a specific platform: "Across the Ocean" (1984) to a model called "Treblo_v3" and "In My Mind" (1982) to ElevenLabs Music, a service launched more than forty years later.
Why it happens and who pays the price
DMN refers to a study, not identified in the article, according to which some detectors rely on incidental characteristics such as sample rate and bitrate, i.e. on properties of the file rather than of the music. In IAM's assessment, the financial damage (blocked accounts, halted releases, withheld royalties) falls mainly on independents; Grafert, quoted by DMN, observes that it cannot be right for emerging bands to have to prove, track by track, that they are real people.
The issue comes at a time of ever-tighter filters: Deezer said in July that in June 2026 AI-generated tracks exceeded half of daily uploads for the first time, around 90,000 a day according to TechCrunch; Bandcamp has banned music generated wholly or largely with AI; and DMN itself reported on 3 November possible side effects of Spotify's anti-spam filters on real musicians. The labelling promoted by IFPI, RIAA and A2IM, by contrast, is based on self-declaration by labels and distributors.
What remains to be clarified
The IAM test is not a published study: the full methodology, accuracy rates for individual tools and responses from the detector makers are not available, and the results are known only through DMN's account. One point, however, is hard to get around: if a detector attributes a 1982 track to ElevenLabs, its verdict cannot be taken as proof. For platforms and distributors that use these scores in moderation decisions, the open question is what human review or appeals procedure follows an "AI" label.



