The Rubberz Problem: Music's AI Fight Is About Disclosure, Not Detection

A synth-pop single credited to Fenix Flexin, a rapper known for trap music as part of the duo Shoreline Mafia, reached No 58 on the Billboard Hot 100 and more than 35m Spotify streams, and this month he conceded he had used AI on it ^{[1][2][3][4][9]}. The useful lesson is not that a machine made a hit. It is that the identification held up while the commercial machinery around it did not pause.

The timeline is the argument. The track came out in June ^[1]. Sceptics alleged AI involvement almost immediately, and Fenix and his credited producer, Purps on the Beat, initially denied it, with Fenix saying he had used Auto-Tune and a fake British accent ^[5][6]. The indie DJ-producer Medasin then posted a video he said showed the song was made with the free generative tool Treblo, and prompted Treblo into producing a track close to it in sound and lyrics ^[7]. Fenix posted what he said were the song's stems; Medasin replied by demonstrating that AI songs can be broken into stems too ^[8]. Then came the admission: "never said I didn't use AI, i said it had nun to do w recording process" ^[9]. That is roughly two months of chart position and streaming revenue accruing before the credited artist stopped denying ^[1].

Detection, meanwhile, was not the bottleneck. Treblo's own open-source classifier rates the track very likely Treblo, according to Wired, and co-founder Ryan Tremblay told the Guardian the score is "higher by far than any false positive" ^[10][12]. The classifier has not been independently reviewed ^[11]. So the most confident public detector belongs to the company whose model is implicated, and its co-founder frames the outcome as a feature: Treblo was built "for everyone, from normal people experimenting at home to charting artists looking for new sounds" ^[13]. Fenix's representative declined to comment ^[14].

The rights-holders' answer arrived on 29 July, before the admission, when Universal, Sony, Warner and others proposed that global charts exclude AI tracks unless they are properly authorised, lawful, substantially human-made, and free of streaming or chart manipulation concerns ^[15][16][2]. Read the criteria closely: authorisation, lawfulness and human authorship are facts about how a record was made and licensed, not properties audible in the file. They can only be established by someone declaring them at the point of delivery. The labels' own stated concern points the same way, that many AI tracks on streaming services were trained on artists' music without consent and are often tied to fraudulently boosted streams ^[17].

The complication is that the same companies are building the supply. Every major label has announced AI music tool partnerships with the likes of Nvidia, Spotify and Stability AI, and BMG has partnered with Suno ^[18][19]. Tyga, a labelmate of Fenix, told Vibe his album $tarface was made with AI ^[20]. This is the first AI-generated single to crack the Hot 100 but not the first AI act on a Billboard chart: in November, Breaking Rust took No 1 on country digital song sales with Walk My Walk ^[22][23].

Watch whether Billboard adopts the criteria and, if so, whether it requires an attestation at delivery rather than an investigation after the fact; whether label AI partnerships carry disclosure metadata into the distribution chain; and whether any classifier not owned by a generator gets independent review.