LA rapper Fenix Flexin appeared to admit to using AI on his viral 80s synth-pop track "Rubberz" in an Instagram comment reply, writing that he "never said he didn't use AI," but that AI "had nothing to do with the recording process" and that he "didn't know what the app was until it was mixed and people started commenting" (The Verge).

That comment sits uneasily alongside his other public statements. Asked directly about AI use by interviewer Kazeem Famuyide on Hot 97's "Mornings With Mero," Fenix Flexin stated there is "no AI on it" (The Verge). Since "Rubberz" was released, he and his team have publicly tried to dispel the AI allegations multiple times, posting claimed clips of recording session files and lip-sync rebuttal clips. In one comment, Fenix Flexin and his team described "Rubberz" as made by two humans recording in his living room with "pro tools auto tune and some reverb" (The Verge).

Fenix Flexin is a member of the Los Angeles rap group Shoreline Mafia (XXL).

Producer Medasin posted a detailed viral video breakdown alleging that "Rubberz" is completely AI-generated (Stereogum). His analysis pointed out that Fenix Flexin left "Sonauto," the former name of the AI tool Treblo, in the song's metadata (Vice). Medasin also demonstrated how easily he could replicate the song's vocal style using Treblo (Stereogum).

Treblo itself entered the fray. The company released a built-in AI detector that identifies "Rubberz" as Treblo AI-created (The Verge). Pitchfork independently tested Treblo's detector on "Rubberz" and reported that it predicts the song's origin is "very likely Treblo," producing similar results for other Fenix Flexin songs (Pitchfork). Wired reported that Treblo's detector also flags songs by Tyga as likely AI-generated (Wired). Tyga, for his part, admitted to relying on AI for his 80s-influenced album "$tarface" (The Verge).

The technical detail that matters here is the metadata trail. Leaving "Sonauto" in a track's embedded metadata is not ambiguous. It is a programmatic signature, the kind of artifact that persists in file containers regardless of downstream mixing or mastering. Whether that signature alone constitutes proof of full AI generation versus partial use in the production pipeline is a distinction Fenix Flexin's Instagram reply seems to gesture at when he separates AI from "the recording process" while not denying use of the tool outright.

That framing, though, does not survive contact with his Hot 97 denial. "No AI on it" is categorical. The Instagram comment is conditional. Those are not reconcilable positions, and the gap between them is where the industry's attribution problem lives.

Treblo's decision to ship a detector that flags its own platform's output adds a wrinkle. The tool's creator now serves as both the supplier of generative capability and the arbiter of its detection. Pitchfork's independent testing producing the same result lends credence to the detector's reliability in this specific case, but the structural conflict is worth noting: a platform that profits from generation also controls the narrative around detection.

The parallel case of Tyga is instructive in one narrow way. Tyga admitted AI use outright, which closed the loop. Fenix Flexin's trajectory has been the opposite: repeated denials, escalating evidence, and finally an Instagram reply that concedes AI involvement without fully owning it. The arc from denial to partial admission is becoming a recognizable pattern as AI-generated or AI-assisted music moves from novelty to controversy to normalization.

The broader context here is that AI music generation tools have reached an inflection point where their output quality is sufficient to produce commercially viable, charting tracks, and the industry's infrastructure for attributing, labeling, and regulating that output has not caught up. Metadata signatures, platform-native detectors, and third-party forensic analysis are each partial mechanisms. None is standardized. None carries regulatory weight. The result is a landscape where artists can deny, hedge, or admit, and where the evidentiary tools available to listeners and press are built by the same companies whose products are in question.

What this enables, longer term, is a more honest accounting. If platform-native detectors become reliable and routine, AI-assisted tracks could be labeled at the point of distribution, the way nutritional information appears on packaging. That outcome depends on the same companies that profit from generation agreeing to transparently flag it. Whether the market rewards that transparency or punishes it is the open question.