Sure Seems Like Fenix Flexin Used AI Music Generator Treblo

Treblo released an open-source audio classifier on August 4, 2026 that detects whether a song was generated by its own platform — and the tool's first headline result is rapper Fenix Flexin's "Rubberz," which Treblo's CEO says may be the first AI-generated track to reach the Billboard Hot 100. The announcement arrives after weeks of public dispute over the song's origins, with musician Medasin's TikTok allegation going viral and Fenix's subsequent denial now itself being called a fabrication.

What Changed

The controversy around "Rubberz" began not with a detection tool but with a stylistic gap. Fenix Flexin built his reputation on West Coast rap, so the song's '80s-influenced pop sheen struck listeners as a sharp departure — pleasant surprise initially, then suspicion. By late July, musician Medasin had posted a TikTok claiming the track was produced entirely with Treblo, a statement that spread widely enough to draw coverage from HotNewHipHop on July 31, 2026, framing it as the moment the debate "may have found an answer" for the general public. Notably, Medasin's allegation went beyond claiming AI involvement in general — he identified the specific tool, which is a more precise and verifiable claim.

Fenix responded by posting what he framed as session files: the kind of project-view screenshot that normally serves as evidence of hands-on production. That rebuttal was quickly unpicked. Medasin and others who analyzed the footage concluded the session was reconstructed after the fact using an AI stem separator — a tool that disassembles a finished audio file into component tracks (vocals, drums, bass) and reimports them into a digital audio workstation, producing an interface that looks like a recording project without any actual creation history behind it. If that analysis holds, the denial is not an alibi. It is a second AI operation designed to obscure the first.

On August 4, Treblo entered the story directly by releasing the Treblo AI Music Classifier as open-source software. According to the company's blog post, the tool returned a "very likely Treblo" confidence score on "Rubberz." Treblo's CEO publicly framed this as the potential first Billboard Hot 100 appearance by an AI-generated song — a milestone multiple outlets, including The Verge and HotNewHipHop, have reported without independent verification from Billboard or any chart authority. Fenix has not publicly addressed the classifier result.

How It Works

The classifier operates on a narrow forensic premise: AI music generation systems leave statistical signatures in their output, patterns shaped by the training data and generation architecture specific to each platform. Treblo's tool was built entirely on audio its own system produced, making it well-calibrated for its own fingerprints and, by design, blind to everything else. It does not detect AI-generated music in general — only audio that Treblo's model specifically created.

The output is a probability score rather than a binary verdict. That distinction matters for anyone thinking about content review tooling: a confidence score communicates calibrated uncertainty, while a yes/no flag hides it. Treblo reports a false positive rate below 1 in 1,000 — meaning fewer than 0.1% of genuinely human-produced tracks would incorrectly register as Treblo-generated. That is a credible threshold for this kind of tool, though it is self-reported and has not been independently replicated by any third party as of publication.

The stem-separator evasion technique alleged against Fenix's rebuttal deserves attention on its own terms. Tools like Demucs and similar open-source audio separators can take any finished audio file and extract isolated component stems. Those stems can then be loaded into a DAW to create a session view that resembles an authentic recording project with no actual production history behind it. Our read is that this technique is not esoteric — it is trivially accessible to anyone who knows it exists — which means session-file screenshots carry essentially no evidentiary weight as "proof of human authorship." Any rebuttal built on that evidence is vulnerable before the conversation even starts.

What It Means for Developers

For anyone building moderation systems, publishing pipelines, or platforms that handle audio at scale, the structural limit here is worth naming plainly: fingerprint-based detection cannot generalize beyond the generator whose outputs trained the classifier. Switch tools and you evade detection entirely. The Treblo classifier is well-built for what it does, but a creator who used Suno, Udio, or any other AI music platform remains invisible to it. This is not a gap that can be patched with more training data — it is the ceiling of the method.

The more durable approach is provenance embedded at the point of creation. Standards like C2PA (Coalition for Content Provenance and Authenticity) use cryptographically signed metadata to record the generating tool, timestamp, and creation parameters in a form that travels with the audio file. That chain of custody does not depend on recognizing statistical fingerprints that might vary across model versions; it documents origin at the source. Adoption in music generation tools specifically remains thin, which means the provenance-at-creation window is open for platforms that want to be ahead of this problem rather than chasing it after the fact.

The PR dynamic in this story is worth noting separately. Treblo released a classifier, applied it to a specific song, and let its CEO claim a Billboard milestone — all without requiring any participation or confirmation from Fenix Flexin. The company absorbs the credibility of a real-world detection event; the artist absorbs the reputational exposure. That asymmetry will repeat as other AI generation platforms face attribution disputes. For developers building generative audio tools, it is worth considering in advance whether releasing a detection product is a liability hedge or a liability transfer — and which side of that trade the platform ends up holding when the first high-profile case arrives.

Sources