Is 'Rubberz' an AI song? The evidence against Fenix Flexin is piling up

A rapper's unlikely synth-pop hit has experts pointing to glitchy drums, impossible vocal range, and lyrics that read like a chatbot wrote them on a deadline.

AI2Day NewsdeskUpdated Editor: Lee Brown4 min read
A bright, modern open-plan workspace photographed from a low angle looking up, with warm natural light coming through large windows
Share

Key points

  • "Rubberz" by Fenix Flexin reached number 58 on the Billboard Hot 100 in 2024.
  • Five AI music detectors each estimated a 20 to 30 percent chance the song is fully or partly AI-generated.
  • The single's cover art and a promotional post were flagged as AI-generated with over 97 percent certainty by multiple image detectors.
  • AutoTune tidies pitch errors; it can't extend a singer's range or hand them a British accent.
  • Fenix Flexin has denied the allegations and posted screenshots of project files, but hasn't released the underlying recordings.

Fenix Flexin built his name as half of Shoreline Mafia, a trap-rap duo from Los Angeles. So when "Rubberz" arrived as a dreamy, synth-driven track soaked in 1980s British pop influence, complete with a fake English accent and a vocal melody that sounds borrowed from Morrissey, listeners noticed. Some liked it. A lot more started asking questions.

What actually sounds wrong with the track?

The recording carries marks that audio experts associate with AI music generators: software that composes and produces music from a text prompt or style instruction.

Songwriter Charlie Harding, co-host of the music analysis podcast Switched on Pop, told The Verge that the backing vocals "duck in and out unexpectedly" and that the drum reverb is "full of unnatural digital artifacting reminiscent of a 64kbps MP3 downloaded off of Napster." He argues this happens because most AI music companies trained their systems on low-quality audio, and that degradation bleeds into the output.

The hi-hats sound brittle. Reverb tails, the natural fade of an echo, cut off abruptly instead of dying away. Chorus vocals carry the muffled texture of a heavily compressed file.

Then there are the lyrics. "Swipin' cards and stackin' chips, I saw you sinkin' ships" raises a simple problem: who's sinking ships? That image never returns. The rhyme scheme is a plain AA, BB pattern, every line ending cleanly, which is exactly what a language model tends to produce when asked to write rap. Experienced rappers use internal rhymes (rhymes inside a line, not just at the end) and near-rhymes to create texture. There are none here.

Producer Curtiss King fed the lyrics into Claude, Anthropic's AI assistant. It flagged multiple signs of AI authorship. Google's Gemini reached the same conclusion.

AI2Day covered a structurally similar case on 13 July 2026, when Josh Fawaz's cover of 'Like a Prayer' hit number one on Australian radio and music experts raised identical red flags about AI production artifacts.

What do the live performances show?

This is where the case gets harder to dismiss. In early in-studio appearances, Fenix lip-synced rather than singing live. When he does sing, he can't match the vocal range on the recording, and in several clips the original studio vocals are audible playing over his live attempt, kept low in the mix.

Fenix says the difference comes down to vocal processing, including AutoTune. Music critic Anthony Fantano and Curtiss King both push back. King compares AutoTune to spellcheck: it tidies errors, it doesn't hand you a new voice. It wouldn't give a West Coast rapper a British accent or extend his range by several notes.

Fenix has posted screenshots of what he says are Pro Tools session files, the digital workspace a producer uses to build a track. Screenshots prove little without the underlying files, and he hasn't shared those.

What does this mean for listeners?

You're not obliged to care whether a song was made by a person or a machine. But it matters for the artists whose work AI systems were trained on without permission. Harding warns that this relatively rough-sounding result still found an audience, and that more capable tools are already producing cleaner, harder-to-detect output.

The uncomfortable part, as our July story on 1010Benja's Suno experiment showed, is that AI-assisted music can be genuinely good. "Rubberz" doesn't clear that bar. What it clears is the detection bar, and that's the thing worth watching.

© 2026 AI2Day