Treblo, the AI music generator formerly known as Sonauto, open-sourced its AI Music Classifier on August 3, 2026, publishing the code so anyone can inspect and test it (Treblo Blog). The classifier is built to detect whether a song was produced using Treblo's own model specifically. According to the company's blog post, the false positive rate is below 1 in 10,000. The Verge, reporting two days later, cited the figure as less than 1 in 1,000 (The Verge).

The distinction matters. The classifier does not detect AI-generated music in general, only music from Treblo's model. Songs made by other AI tools would not trigger a positive result. Treblo maintains a GitHub organization account with a single public repository (GitHub) where the classifier's code is available.

The release landed in the middle of an active controversy. Fenix Flexin, an LA-based rapper, released a single called "Rubberz" that faced AI-generation allegations almost immediately (Pitchfork). Wired reported that hip-hop fans had been arguing over whether the track was machine-made, with some claiming to have proof (Wired). Futurism reported that a producer called the song out as AI-generated, prompting Fenix Flexin to deny the allegations (Futurism).

Treblo's classifier weighed in. When run against "Rubberz," the tool classified the track as "very likely Treblo" with "high confidence" (The Verge). Treblo CEO Ryan Tremblay said in a statement to The Verge that the classifier returned a very high probability score, indicating the released audio was generated almost entirely by Treblo's model. If confirmed, that would make "Rubberz" the first known AI-generated song to reach the Billboard Hot 100 (The Verge).

Fenix Flexin denied that "Rubberz" was made using AI and posted clips of what he claims are project files from the recording session (The Verge). The producer Medasin contended that Fenix Flexin used an AI stem separator, a tool that splits a finished song into individual instrument and vocal tracks, to fake the recording session shown in those project-file videos (The Verge; Stereogum). Fenix Flexin did not immediately return a request for comment from The Verge (The Verge).

The false positive rate discrepancy between Treblo's blog (below 1 in 10,000) and The Verge's reporting (less than 1 in 1,000) is worth noting. The blog post, published August 3, is the earlier source. Whether The Verge received an updated figure from Treblo or is citing a more conservative number is not clear from the reporting. For a tool whose output is being used to publicly contest a charting artist's authorship, the difference between those two thresholds is not trivial. At 1 in 10,000 false positives, roughly one song in ten thousand flagged by the classifier would be a false alarm. At 1 in 1,000, that rate is ten times higher.

The broader context here is a structural tension in deploying a vendor-built, vendor-specific detection tool as an arbiter of provenance. The classifier detects Treblo output because Treblo has privileged access to its own model's artifacts, training data, and generation pipeline. That access gives the tool strong signal for Treblo-produced audio but tells us nothing about the broader landscape of AI music generation. A negative result does not mean a song is human-made; it means the song was not detected as Treblo-made. That distinction will matter more as additional AI music tools enter the market and as detection becomes a factor in copyright disputes, chart eligibility, and platform moderation.

The open-source release does at least allow independent verification of the classifier's methodology. Third parties can inspect the code, run it against their own datasets, and test the false positive rate claims empirically. That is a meaningfully different posture from closed-source provenance assertions, where the only recourse is to trust the vendor's stated metrics.

The "Rubberz" dispute may resolve itself, or it may linger as an unresolved question in the way that provenance disputes tend to when the evidence is probabilistic rather than binary. What is clear is that the tools for making these claims are now public, vendor-aligned, and operating in a space where the stakes include chart positions and professional reputations.