Music Origin Report: Beyond “AI-assisted”

“AI-assisted” tells us that a tool was used. It does not tell us whether the recording contains a human performance, generated vocals, generated accompaniment, or a mixture of sources. Authio’s Music Origin Report: Beyond “AI-assisted” examines that distinction across 79,386 successful recorded analyses.

Among the 35,889 analyses that retained a complete classification, 15.03% were Hybrid and 8.89% were AI. Hybrid classifications outnumbered AI classifications by 1.69 to one. These are observations about material analysed by Authio, not an estimate of the share of AI music released worldwide.

Download the Music Origin Report (PDF, 11 pages)
Authio Research · Research edition 1.0 · 16 September 2026

Human, Hybrid and AI: what the classifications show

Authio’s September 2026 report records 27,288 Human, 5,395 Hybrid, 3,189 AI and 17 Inconclusive outcomes in its complete-classification population. This population covers 45.21% of the 79,386 recorded analyses. The percentages below use the 35,889 complete classifications as their denominator.

Recorded outcomes among 35,889 complete classifications

Classification	Analyses	Share
Human	27,288	76.03%
Hybrid	5,395	15.03%
AI	3,189	8.89%
Inconclusive	17	0.05%

Human, Hybrid and AI are the categories preserved in these records. Historical categories do not automatically establish conformance with the later Authio Origin Classification Standard (AOCS). AI and Hybrid together account for 23.92% of complete classifications, but keeping the categories separate preserves the distinction the report investigates.

Why hybrid music changes the question

A human vocal over generated accompaniment and a generated vocal over human accompaniment can both be described as “AI-assisted”. That description conceals a material difference in the origin of the recorded sound. A human performance processed with an AI tool raises a different question again. These examples illustrate production possibilities; they are not reconstructed cases from the report’s dataset.

For distributors, labels and platforms, the practical question is which audio sources are present and what evidence supports that conclusion. A policy that distinguishes generated performances from processing needs more information than a declaration that AI was used somewhere in the workflow.

The report illustrates the reporting effect in a linked API cohort of 13,210 analyses. Its recorded split is 75.14% Human, 16.98% Hybrid and 7.88% AI. If Hybrid were folded into Human, the displayed Human share would become 92.12%. The audio would not change. The reporting convention would remove 16.98 percentage points of visible hybrid activity. This is an arithmetic illustration, not a measured comparison with another detector.

AOCS v1.0 defines five technical classes: Human-Classified, Human-Dominant Hybrid, Balanced Hybrid, AI-Dominant Hybrid and AI-Classified. Review Required is a routing status. The standard separates technical origin classification from authorship and ownership; an AI score is not a percentage of human labour or creative contribution.

What can be said across all 79,386 analyses?

Across the full report population, 19,931 analyses have either an explicit AI classification or a historical AI detection, representing 25.11% of all 79,386 analyses. A further 5,395 have an explicit Hybrid classification. Because some historical results preserve only a binary detection outcome, the full population cannot support a complete Human/Hybrid/AI split.

Evidence coverage across all 79,386 recorded analyses

Recorded evidence	Analyses	Share
AI classification or historical AI detection	19,931	25.11%
Explicit Hybrid	5,395	6.80%
Explicit Human	27,288	34.37%
Inconclusive	17	0.02%
Unresolved classification coverage	26,755	33.70%

These mutually exclusive evidence groups account for the full population. The unresolved share cannot be assigned to Human merely because a complete AI or Hybrid classification is absent.

The public checker: 48.09% historical AI detections

Authio’s public checker recorded 13,454 historical AI detections among 27,974 successful results, a share of 48.09%. People chose to submit this material for verification, so this is a view of submitted audio rather than a random sample of released music. The remaining 51.91% cannot be labelled Human from that binary history.

To examine sensitivity to repeat activity, the report also retains only the first successful result per browser fingerprint. The AI detection share is then 51.06%, or 6,630 of 12,984 results. Browser fingerprints are not unique people or unique songs, so this check does not establish a listener, customer or track count.

Artifact washing requires evidence beyond repeat submissions

Artifact washing concerns attempts to obscure evidence of generated audio through processing. Detecting repeated submissions alone does not establish that it happened. A change in detector output, a missing platform attribution or similar file metadata cannot establish shared audio identity, the cause of a change or the submitter’s intent.

Authio’s ArtifactWatch research concerns the reliability of origin evidence after processing. This edition does not establish a validated artifact-washing rate or a validated washing experiment. It does not classify repeat activity as fraud. A defensible prevalence claim would require verified audio relationships and evidence that separates ordinary processing from concealment.

Scope, methodology and source

The report covers successful recorded analyses before 16 September 2026 at 07:34:04 UTC. Its reconciled population contains 43,785 API analyses, 7,627 dashboard analyses and 27,974 public-checker results. Repeated submissions, cached responses and internal activity are included. The total measures recorded analyses, not unique songs, customers or fresh model executions.

The report documents record reconciliation, classification precedence, historical linkage validation and exclusions. The 19,931 AI-evidence count combines 3,189 complete AI classifications, 13,454 checker detections and 3,288 historical API and dashboard detections without counting the same reconciled analysis in both groups. Record-linkage validation measures the reliability of matching records; it is not a detector-accuracy benchmark.

Authio develops the forensic detector and produced this observational report. The study is not an independent audit or a representative estimate of the music market. Its contribution is to document what the retained evidence supports, including the substantial hybrid category that a binary account would obscure.

Cite the report: Authio Research. Music Origin Report: Beyond “AI-assisted”. Research edition 1.0. 16 September 2026. Full report and methodology (PDF).

For research enquiries, contact Authio. For the classification framework, see the AOCS Technical Basis and Validation Note.