Is Fenix Flexin's Billboard Hot 100 Hit "Rubberz" AI Slop?

Fenix Flexin's "Rubberz" sits at number 58 on the Billboard Hot 100 — but whether a human being actually wrote or sang it is, as of this week, genuinely unresolved. The Shoreline Mafia rapper has denied using AI tools, posted supposed evidence, and then deleted it within hours, as a growing stack of expert analysis picks apart the recording's artifacts. It may be the first track suspected of being largely or entirely AI-generated to chart at this commercial scale.

Background You Need

Fenix Flexin made his name as one half of Shoreline Mafia, a Los Angeles rap duo whose sound runs trap-inflected and unmistakably West Coast. "Rubberz" is something else entirely: melodically, it borrows from 1980s British synth pop, with Fenix adopting a vocal style that multiple listeners have compared directly to Morrissey. That kind of stylistic leap isn't by itself proof of anything — artists cross genres, and novelty isn't a tell. But the combination of that radical departure and very specific sonic anomalies inside the recording is what brought serious scrutiny.

The scrutiny arrived publicly through artist and producer Medasin, who published a detailed breakdown claiming "Rubberz" is fully AI-generated and naming a specific tool: Treblo, an AI music generation platform that was previously called Sonauto AI. Medasin's central piece of circumstantial evidence was Fenix's own prior behavior — before "Rubberz" was released, he had posted a song snippet that carried the label "Sonauto" in its filename. Adding an odd timestamp to the story: Sonauto AI rebranded as Treblo exactly two days before "Rubberz" dropped. Medasin then demonstrated the software generating lyrical phrases and melodic material specifically similar to what appears in the track.

For his part, Purp On The Beat, the credited producer, sidestepped direct denials and argued instead that AI should be treated like any other production tool — and that Auto-Tune is "basically AI." Will.i.am, who has been vocal about AI music adoption, called the song "dope" while also noting what he described as obvious AI involvement. That framing — neither a confession nor a denial, but a philosophical pivot — has become the production side's primary public posture.

What's New

The debate crested when Fenix posted a carousel of video evidence on Instagram: Pro Tools sessions and what he characterized as stems proving the song was assembled through conventional recording methods. Within hours, he deleted the post. Medasin's response was immediate and specific: the sessions showed Fenix taking Treblo's output and running it through a stem separator, a tool that breaks a mixed track into component parts — vocals, drums, bass, instruments. The resulting stems look like traditional production artifacts, but their origin is still AI-generated audio. The alleged receipts were a layer of surface plausibility over the same underlying material.

Charlie Harding, songwriter and co-host of the Switched on Pop podcast, offered the most technically rigorous public analysis of the recording itself, speaking to The Verge. He named five specific artifact types: hi-hats with an unusually brittle texture; chorus vocals that carry the distortion signature of a 64kbps MP3 — his description, echoed across multiple reports, was that it sounded like audio downloaded from Napster; reverb tails that cut off abruptly rather than decaying naturally; backing vocals that appear and disappear without dynamic logic; and syllable stress landing on the wrong beats, a pattern that emerges when a model optimizes for rhyme sound over natural speech rhythm.

That last artifact is one that audio AI detection tools largely miss. Five separate AI music detection platforms returned probability scores in the 20–30% range — not high enough to be meaningful. The contrast with image detection is striking: the cover art for "Rubberz" flagged at 97% or above on AI image detectors, a discrepancy that reveals how far behind audio verification tools have fallen relative to the rest of the detection ecosystem. Text-based tools, including Claude, flagged the lyrics. The lyrical structure Harding and others identified is consistent with what language models default to under pressure: a strict AA/BB end-rhyme scheme with no internal or near rhymes, and lines that follow rhyme logic over semantic logic. One analysis specifically highlighted an unresolved reference — a phrase about "sinking ships" with no antecedent — as evidence of a model completing a sound pattern rather than a thought.

The live performance record compounds the case. Auto-Tune can compress pitch variation, but it cannot manufacture a vocal range or regional accent that wasn't there to begin with. Witnesses to Fenix's live appearances have noted that his voice doesn't cover the same register or stylistic territory the recording does — a discrepancy that experts say rules out studio processing as a full explanation.

The Pushback

Fenix's position isn't incoherent — it's just unverifiable. His counter-narrative holds that the stylistic shift is artistic experimentation, that producers are entitled to use available tools, and that the Pro Tools sessions he briefly shared prove traditional construction. The problem with that last argument, as Medasin documented, is that project files prove assembly, not origin. A stem separator turns a single AI-generated audio file into what looks like a multi-track session. Without access to the original recordings — which Fenix has not shared — the session file proves nothing about where the audio came from.

Our read is that the more significant story here isn't whether "Rubberz" specifically used Treblo. It's that five commercial AI audio detection tools couldn't produce a number above 30% on a track that a credentialed music analyst can annotate artifact by artifact. Harding himself noted that the model quality behind "Rubberz" appears to be a generation behind the current state of the art — better tools have already smoothed out these particular fingerprints. A lower-quality AI output charted at number 58 on the Hot 100 while detection infrastructure returned an inconclusive shrug. Generation has outrun verification, and the gap is only widening.

Sources