Treblo’s Own AI Detector Is Now Flagging Fenix Flexin’s “Rubberz”

The AI music app at the center of a summer-long controversy has effectively turned on its own users. Treblo, a generative AI music tool formerly named Sonauto, now has a built-in detector that flags both "Rubberz" by Fenix Flexin and several tracks by Tyga as likely AI-generated. The song, released June 5, currently sits at No. 58 on the Billboard Hot 100 and has 7 million views on its music video. Fenix and his credited producer have denied using AI throughout, but a video posted by producer Medasin last Thursday laid out a detailed case pointing directly at Treblo.

What happened

Detail	Fact
Song	“Rubberz” by Fenix Flexin, released June 5
Chart position	No. 58 on Billboard Hot 100
Music video views	7 million
AI app implicated	Treblo (previously named Sonauto)
Name change timing	Two days before “Rubberz” was released
Treblo’s verdict	Built-in detector now flags both Fenix and Tyga tracks as likely AI-generated

“Rubberz” raised eyebrows from day one. Fenix Flexin built his reputation as part of Shoreline Mafia in the late 2010s, known for a distinctly nasally, East Hollywood-inflected voice. The voice on “Rubberz” sounds nothing like that. It is brighter, crisper, and according to Wired, closer in accent to “a very chipper Morrissey.” The song is also not rap. It is a synth-pop track that sounds at home in an ’80s playlist, a significant genre shift for the artist.

Fenix addressed the accusations directly. In comments on an On the Radar performance, he wrote: “No sirrrrrrr, recorded same as all music I do! Only difference is auto-tune reverb and me using my fake uk accent lol.” He told Channel 5 he “freestyled” the lyrics. Producer Purps on the Beat has also denied AI involvement.

How Medasin built the case

On Thursday, producer Medasin posted a video laying out the accusation step by step. The key pieces of evidence:

• A filename visible in an old Fenix Instagram Story ends with “Sonauto.mp3,” the former name of Treblo.
• Sonauto was renamed Treblo two days before “Rubberz” dropped on June 5.
• Medasin used Treblo himself with tags like “80s synthwave” and a prompt for “rapper lifestyle” lyrics, generating a song with striking lyrical similarities.

The lyrical overlaps are specific. Medasin’s generated output included “Diamond chains hanging heavy.” “Rubberz” has “Now I bought a heavy diamond chain.” Medasin’s output also gave him “I see your face in the champagne bubbles.” The actual song has “Still seeing your face in the glass.” Both songs use some form of “pockets thick” and both mention “plastic smiles.”

“There’s just way too many coincidences,” Medasin told Wired. “It is just a mathematical impossibility that he did not use Treblo.”

Wired’s own reporter tested Treblo independently. Using “80s synthwave” tags and a rap lyrics prompt, the first attempt produced the couplet “Rings on every finger / Leather seats and cedar scent linger,” compared to the “Rubberz” line “Gold rings on my fingers / Still your perfume lingers.” Repeated tests with slight variations produced similar lyrical structures every time. The app appears to have a strong bias toward finger/linger rhyme pairs, references to rings and glass, and a night-to-day thematic structure.

The app’s own detector sealed it

The most striking development is that Treblo itself added a built-in AI detection feature. That detector now flags the songs connected to both Fenix Flexin and his labelmate Tyga as likely AI-generated. Medasin told Wired he had identified the Treblo connection about a month before posting, but held off expecting a journalist to cover it first. A site called Zinstrel ran a smaller piece, but it did not gain traction until Medasin’s video.

Fenix responded that same Thursday afternoon, posting a video on Instagram purporting to show the original project file for “Rubberz.” An authentic project file would suggest the song was built from scratch in a studio. However, multiple music producers commenting on the post claimed the file looked more like a stem separation of a finished track (a process where software splits a completed song into basic components like bass and drums) rather than a genuine session recording.

Why it matters beyond the drama

This story is a useful case study for anyone thinking about AI detection in creative work. The well-known AI music detectors did not flag “Rubberz” all summer. It took a producer manually reverse-engineering the workflow, finding a filename artifact in a social media post, and recreating the output to make the argument stick. Even then, the clearest confirmation came from the tool itself.

For businesses using AI-generated content, this points to a real operational risk: the metadata and filenames attached to AI outputs can expose the workflow even when the content passes surface-level checks. If you are using AI tools in any creative production, from music to copy to images, it is worth auditing what traces those tools leave behind. Our AI integration work for clients always includes a review of what gets embedded in outputs before anything goes public.

The broader pattern matters too. As we covered in our piece on Firefly vs Midjourney and legal risk in AI art, the question of provenance in AI-generated creative work is becoming harder to sidestep. The tools are getting better, but so are the detectors, and sometimes the most damning evidence is a filename you forgot to rename.

Our take

The specific detail that a filename ending in “Sonauto.mp3” appeared in an Instagram Story is the kind of thing that feels obvious in hindsight and almost impossible to catch in advance. No AI detector flagged this song for months. It was a producer doing old-fashioned forensic work who cracked it open. That should recalibrate how much trust you put in automated AI detection tools right now. They are useful, but they are not comprehensive.

The fact that Treblo’s own detector then confirmed the finding is genuinely unusual. An AI company flagging its own users’ outputs is not a common occurrence, and it raises a question worth watching: what liability, if any, does a generative AI music platform carry when its outputs are passed off as original recordings on commercial releases?

The practical takeaway for anyone working with AI-generated creative assets is simple: treat the output files the same way you would treat source code. Know what metadata is embedded, rename everything before it leaves your workflow, and document your process. The alibi Fenix tried to offer, a project file, was immediately questioned by producers. A clean paper trail from the start is harder to dispute.

Source: WIRED · AI