Spotify signs Merlin remix deal as Treblo releases a model-specific detector

Research report date: 4 August 2026

DAOrecords Signal publication: 4 August 2026 at 14:59:51 UTC

Coverage status: One strict-window record missed by the scheduled scan and one material post-cutoff development found during the extended same-day search

Linked records: DAOR-SIGNAL-20260804-001, DAOR-SIGNAL-20260804-002

Factual reporting

Treblo publishes a model-specific music detector

Treblo, formerly known as Sonauto, released a public classifier designed to estimate whether an audio file was created with Treblo.

The release is anchored to the first commit in Treblo's public classifier repository:

• Commit: 2b18a667ba17bebed718d44b7d2f500d964837ec
• Commit message: Release
• Commit time: 2026-08-03T05:36:26Z
• Licence: GNU Affero General Public License version 3
• Repository: Sonauto/treblo-classifier

The exact commit time falls inside the scheduled scan's window of 2026-08-03T04:02:18Z through 2026-08-04T04:02:18Z.

The classifier can be run locally from the command line. It accepts one or more audio files and returns a Treblo probability, a label and a confidence level. Its first run downloads the treblo-classifier-v1.ckpt checkpoint from Treblo's Hugging Face repository and caches it for later use.

Treblo also provides a browser-based detector. Its public interface states that uploaded files are temporary and deleted after analysis.

Declared limits

Treblo explicitly states that the classifier is not a general AI-music detector.

It is tuned to identify songs made with Treblo with little or no modification and will not identify music generated by other platforms.

The official documentation also states that:

• Treblo-generated audio can be modified to appear non-Treblo more easily than non-Treblo audio can be made to appear Treblo;
• extra audio can affect the result;
• higher-quality WAV or FLAC input is more accurate than low-bitrate compressed audio;
• the probability, label and confidence remain estimates rather than a complete provenance record.

Public origin dispute

WIRED reported that the detector launched while the origin of commercially successful releases by Fenix Flexin and Tyga was being publicly disputed.

The article states that Treblo's detector returned Very likely Treblo for Fenix Flexin's “Rubberz” and Likely Treblo for a Tyga track. It also reports that Fenix Flexin and the credited producer had denied using AI, and that requests for further comment were unanswered.

Treblo reportedly said it had not previously spoken with the artists and had not known about “Rubberz” until after it entered the Billboard Hot 100.

Treblo claims a false-positive rate below one in 10,000. That performance claim has not been independently reviewed in the cited materials.

DAOrecords does not treat these detector results or surrounding commentary as proof that either artist used Treblo. The origins of the named releases remain disputed unless stronger evidence, an admission or an authoritative determination establishes otherwise.

Spotify signs a Merlin agreement for its fan-made covers and remixes tool

Reuters reported at 2026-08-04T10:11:00Z that Spotify announced a new agreement with Merlin for Spotify's upcoming paid tool for fan-made covers and remixes.

Reuters states that the agreement will allow artists on labels participating in Merlin's Spotify agreement to take part.

The announcement occurred after the scheduled scan's 2026-08-04T04:02:18Z cutoff and was found during the extended same-day search.

Spotify previously described the tool in its May 2026 agreements with Universal Music Group and Universal Music Publishing Group. Spotify said the paid add-on for Premium users would use generative AI to let fans create covers and remixes from participating catalogues, with consent, credit and compensation built into the product.

The Merlin agreement is a separate product-specific expansion beyond the earlier UMG agreements. It potentially brings independent-label repertoire represented through Merlin into the participation framework.

No product-specific Spotify or Merlin announcement was located during this research run. The exact Merlin agreement, operative participation terms and rights coverage were therefore not independently reviewed.

Why it matters to the music industry

The two developments address opposite ends of the same provenance and participation chain.

Treblo's classifier concerns attribution after audio has been created. Spotify's Merlin agreement concerns permission, participation and compensation before fan-made derivative outputs are enabled.

Provider-specific attribution

A general detector attempts to infer whether audio contains characteristics associated with machine generation across multiple systems. A provider-specific classifier searches for characteristics associated with one model family and may therefore produce a narrower, more actionable result.

The public release of Treblo's code changes the evidence environment:

• labels, distributors, DSPs and researchers can inspect the implementation;
• a detector can be run locally without submitting unreleased audio to a third-party web form;
• results can be reproduced against a defined checkpoint and tool version;
• independent researchers can test false positives, false negatives, compression sensitivity and evasion;
• providers can be asked to publish model-specific detectors alongside generation systems.

The accompanying dispute shows the limit: a probabilistic result may conflict with artist or producer statements and must not be converted automatically into a public accusation.

Licensed fan participation for independent repertoire

Spotify's agreement with Merlin expands the commercial and governance relevance of its planned covers and remixes product beyond a major-label catalogue.

For independent artists and labels, participation may affect:

• master-recording authority;
• composition and publishing authority;
• featured and non-featured performer interests;
• consent and opt-in scope;
• approved transformation types;
• attribution and metadata;
• derivative-output ownership and permitted use;
• royalty calculation and allocation;
• platform pricing and revenue share;
• territory and term;
• withdrawal, takedown and post-termination treatment;
• model training or retention restrictions;
• public release and export rights.

Merlin's existing Spotify relationship does not by itself establish that every Merlin member, artist, recording or composition is automatically included. Participation must remain tied to the actual product agreement and valid artist, label, publisher and performer authority.

DAOrecords analysis

Add model-specific detection to the evidence model

The Release AI Origin and Monetization Profile should distinguish:

• general AI detector;
• provider-specific classifier;
• embedded watermark detector;
• metadata or content-credential validator;
• platform-applied label;
• artist or contributor declaration;
• provider account or generation record;
• technical forensic review.

For every detector run, DAOrecords should preserve:

• detector provider and product;
• detector type;
• source-code version or commit;
• model checkpoint and hash where available;
• licence;
• execution environment;
• local or remote processing;
• input file hash and format;
• audio duration and quality;
• known edits, mastering, compression or added audio;
• probability, label and confidence;
• analysis time;
• operator;
• declared limitations;
• supporting and conflicting evidence;
• review and correction history.

A detector output should never overwrite the artist-confirmed creation record. It should become a separate evidence item that may support, conflict with or leave the release-origin state unresolved.

Evaluate local detector deployment as a governed connector

The open-source Treblo CLI is relevant to the DAOrecords Music Data Connector Evaluation because it can be tested locally.

Before adoption, the connector-evaluation process should assess:

• code and dependency security;
• checkpoint provenance and integrity;
• repeatability across file formats;
• false-positive and false-negative rates;
• resistance to mastering, compression, stem processing and adversarial edits;
• resource requirements;
• update and supersession policy;
• output schema;
• privacy and retention;
• licence obligations;
• whether results can be independently reproduced.

The browser detector's deletion statement is a provider claim. DAOrecords should not upload unreleased Child assets without a separately approved privacy, terms and evidence review.

Treat Spotify participation as a governed rights package

The Rights-Compliant Remix and Fan Participation workflow should model the Spotify–Merlin product as an opt-in participation package rather than infer access from catalogue delivery.

Before a Child release or catalogue asset participates, DAOrecords should preserve:

• the participating Merlin member and applicable Spotify agreement;
• controlled master and composition shares;
• songwriter, publisher and performer permissions;
• eligible recordings, territories and versions;
• permitted cover and remix operations;
• prohibited uses;
• AI provider and model where disclosed;
• source-asset access;
• attribution and disclosure;
• derivative-output ownership and licence;
• revenue formula, deductions and split rules;
• artist approval;
• withdrawal and takedown procedure;
• output availability after withdrawal;
• evidence and agreement versions.

The current report does not establish those terms. Their absence is a review requirement, not permission.

Preserve disputed-origin safeguards

Where a classifier conflicts with an artist, producer or contributor statement:

1. preserve the original file and its hash;
2. preserve the detector version, checkpoint and complete output;
3. obtain relevant project files, stems, exports or generation records where authorized;
4. preserve all contributor statements;
5. run an independent or second-method review where proportionate;
6. record the state as disputed or unresolved;
7. avoid public accusation or automatic release blocking without applicable authority;
8. update the record transparently if stronger evidence later emerges.

Provider-specific detection can improve provenance evidence. It cannot determine authorship, copyright ownership, training authorization, voice consent, contractual permission, disclosure compliance, chart eligibility or legal liability by itself.

Assessment

DAOR-SIGNAL-20260804-001

• Event: Treblo releases an open-source model-specific music detector
• Impact level: High
• Confidence: High that the detector and source code were released; detector efficacy remains independently unverified
• Source status: Confirmed product and code release
• Affected components: Release AI Origin and Monetization Profile; AI Rights Profile; Music Data Connector Evaluation; Rights + Metadata; ReleaseOps; Evidence Preservation; Ecosystem Relations
• Canonical source: Treblo classifier release commit (opens in a new tab)
• Official detector: Was this made with Treblo? (opens in a new tab)
• Supporting report: WIRED investigation (opens in a new tab)
• Canonical publication time: 3 August 2026 at 05:36:26 UTC
• Commercial data value: High
• Primary limitations: Model-specific rather than general; modifications and input quality can affect results; reported false-positive performance is not independently validated; named release origins remain disputed; detection does not establish legal or contractual provenance.

DAOR-SIGNAL-20260804-002

• Event: Spotify announces a Merlin agreement for its paid fan-made covers and remixes tool
• Impact level: High
• Confidence: High that the agreement was announced; lower regarding undisclosed product and participation mechanics
• Source status: Confirmed agreement
• Affected components: Rights-Compliant Remix and Fan Participation; AI Platform Programme and Sponsorship Review; Release AI Origin and Monetization Profile; AI Rights Profile; Rights + Metadata; Finance + Royalties; BizDev; Ecosystem Relations
• Canonical source: Reuters report (opens in a new tab)
• Product context: Spotify–UMG fan-made covers and remixes announcement (opens in a new tab)
• Canonical publication time: 4 August 2026 at 10:11:00 UTC
• Commercial data value: High
• Primary limitations: No product-specific Spotify or Merlin announcement was located; operative agreement terms, member opt-in, launch timing, price, territories, rights coverage, compensation formula, output rights, withdrawal and post-termination treatment remain undisclosed.

Daily synthesis

AI-music infrastructure is developing simultaneously around attribution and permission.

Treblo released a reproducible provider-specific detector that can strengthen technical evidence while remaining vulnerable to modification, input quality and unresolved false-positive questions.

Spotify's Merlin agreement broadens the potential independent-label participation layer for licensed AI-enabled covers and remixes, but the operative rights and compensation mechanics remain unavailable.

For DAOrecords, neither a detector score nor a platform agreement should be collapsed into a binary AI status. The system needs a dated evidence chain that records model attribution, source authority, rights scope, participant consent, transformation permissions, output terms, compensation and unresolved conflicts.

Record index