Fenix Flexin's Billboard hit "Rubberz" looks like AI-generated music
Fenix Flexin, a member of the Los Angeles rap duo Shoreline Mafia, released the solo track "Rubberz," which climbed to number 58 on the Billboard Hot 100. Almost immediately, listeners and commentators speculated the song was largely or entirely generated by AI. Fenix has denied the accusations but done little to dispel them.
The suspicion starts with a stylistic break: Fenix has built his career on trap-influenced West Coast rap, while "Rubberz" is built on '80s UK synth pop, with a vocal melody drawing on Morrissey and Fenix affecting a fake British accent. That alone proves nothing, since artists change styles.
More telling, according to the piece, is the recording itself. Close listening reveals artifacts commonly associated with AI-generated music: brittle-sounding hi-hats, chorus vocals that resemble a low-bitrate MP3, and reverb tails that cut off unnaturally. Charlie Harding, a songwriter and co-host of the Switched on Pop podcast, told The Verge: "Ghostly backing vocals duck in and out unexpectedly. The drum reverb is full of unnatural digital artifacting reminiscent of a 64kbps MP3 downloaded off of Napster. Because most generative AI companies have trained their model on low-quality MP3s of unlicensed copyrighted material, we hear audibly degraded instrumentation."
Running the track itself through five different AI-music detector tools did not settle the question: all five put the chance the track is AI or AI-assisted at between 20 and 30 percent, a range the author calls short of a smoking gun, noting such tools lag behind the music-generation models they are meant to catch. Supporting materials fared worse: the single's cover art and a social post celebrating its chart success were both flagged as AI-generated with over 97 percent certainty by multiple AI image detectors. The lyrics, fed into multiple AI writing detectors and into Gemini, were judged by all of them to read as AI-written; producer and YouTuber Curtiss King ran the same lyrics through Claude, which likewise flagged multiple signs of AI generation.
The lyrics themselves read as logical near-gibberish: simple, clean rhymes in a basic AA-BB scheme, with no near rhymes or internal rhymes, and lines that fail to connect into a coherent narrative (a line about "sinking ships" that is never mentioned again; a chorus about being "able" to do something that is never named). Harding also flagged the delivery, calling it a "complete lack of flow" and singling out the title line, "I just let the rubber band snap," for syncopation that "emphasizing all the wrong syllables" by forcing the word "the" onto the downbeat instead of "rubber."
Live performances have not helped Fenix's case. Early TV appearances showed him lip-syncing awkwardly, seemingly unable to remember his own lyrics or flow. In performances where he does sing live, his voice cannot reproduce the recording's vocal range or accent, and in several clips the original studio vocals are audibly played over his live ones, with his real voice mixed far down and the backing track occasionally dropped out in an apparent, unsuccessful attempt to prove authenticity.
Fenix has attributed the gap between the studio and live versions to vocal processing, including AutoTune. Harding, King and music critic Anthony Fantano have all disputed that explanation; Fantano's argument, echoed by King's comparison of AutoTune to spellcheck, is that the tool corrects pitch errors but does not extend a singer's vocal range or manufacture an accent. Fenix has also posted clips described as Pro Tools session files, which the author says prove nothing without the original recording and project files being shared publicly, something the piece concludes is unlikely to happen. Harding adds a final wrinkle: if the track is AI-made, it was made with an inferior model, since more advanced AI music tools have already reduced the kind of audible artifacts and rhythmic awkwardness heard on "Rubberz."
Key facts
- Fenix Flexin's track "Rubberz" climbed to number 58 on the Billboard Hot 100 and is suspected of being largely or entirely AI-generated.
- Five separate AI music detectors each put the chance the track is AI or AI-assisted at between 20 and 30 percent; the single's cover art and a celebratory post were separately flagged as AI-generated with over 97 percent certainty by image detectors.
- Lyrics fed into multiple AI writing detectors, Gemini and Claude were all judged to show signs of AI generation, and the lyrics themselves read as rhyme-driven near-gibberish lacking narrative coherence.
- Songwriter Charlie Harding identifies specific audio artifacts (low-bitrate-style digital noise, unnatural reverb cutoffs, off-syllable syncopation) typical of AI-generated music.
- Live performances undercut Fenix's AutoTune defense: his live voice cannot match the recording's range or accent, and original studio vocals are audibly used in place of his live singing in several clips.
Why it matters
A song suspected of being substantially AI-generated reached number 58 on the Billboard Hot 100, one of music's most closely watched charts, without the artist confirming or denying AI involvement in a way that settles the question. It is a concrete, mainstream case of AI-made or AI-assisted content passing as a human artist's work in a commercial context that pays out on streams, airplay and chart position, rather than a hypothetical about AI music's future.
Who it affects
Listeners and the artist's existing fanbase, who are being asked to evaluate authenticity with the tools available to any journalist rather than any special access; other musicians and labels, who now face comparison against AI-made tracks on the same charts; Billboard and streaming platforms, whose chart and payout systems currently make no distinction between human and AI-made music; and the makers of AI detection tools, whose products are shown here to agree with each other and fall well short of certainty.
How to use it
The piece models an approach anyone can apply to a suspect track: check for a genre or style discontinuity that breaks from the artist's known catalog, listen for a specific list of audio artifacts (brittle hi-hats, low-bitrate-sounding vocals, reverb tails that cut off unnaturally, syncopation that lands on the wrong syllables), examine the lyrics for a rhyme scheme that prioritizes rhyme over sense, run the audio and lyrics through more than one AI detector rather than trusting a single tool, and weigh how the artist performs the song live against the studio recording.
How solid is it
The evidence is circumstantial rather than conclusive, and the piece is explicit about that: the five AI music detectors run on the track itself returned only a 20 to 30 percent AI probability, not a confirmation. The stronger signals sit around the track rather than in it, the cover art and a promotional post were flagged as AI-generated with over 97 percent certainty by multiple image detectors, and the lyrics were judged AI-written by every writing detector and chatbot tested, including Gemini and Claude. Independent commentary from a professional songwriter, a producer and a music critic converges on the same read of the audio artifacts and the live-performance discrepancies. The author's own verdict is stated plainly: "We can't be 100 percent sure, but probably."
Risks and caveats
AI music and AI content detectors are acknowledged in the piece to be unreliable and to lag behind the generative models they are meant to catch, so a 20 to 30 percent AI-probability reading on the track itself is not proof either way. Fenix Flexin has denied the accusations and attributes the gap between his studio and live performances to vocal processing including AutoTune, an explanation disputed by Harding, Curtiss King and Anthony Fantano but not independently disproven with technical evidence beyond their stated arguments. No AI tool, platform or company is named as having been used, no one from Billboard, Fenix's label or Shoreline Mafia is quoted, and the source itself frames its conclusion as a probability rather than a confirmed fact.
“Ghostly backing vocals duck in and out unexpectedly. The drum reverb is full of unnatural digital artifacting reminiscent of a 64kbps MP3 downloaded off of Napster. Because most generative AI companies have trained their model on low-quality MP3s of unlicensed copyrighted material, we hear audibly degraded instrumentation.”
— Charlie Harding, songwriter and co-host of the Switched on Pop podcast