Every major music platform now claims to label AI-generated content. TikTok's automatic system gets it right about 30% of the time.
That number comes from Dr. Neal Krawetz, a forensic analyst who has spent years building detection tools for manipulated media. In a detailed analysis published this week, Krawetz argues that the music industry's approach to AI content is broken at every level - and that the same failures are playing out across photos, video, and text.
The Binary That Doesn't Exist
The core problem: there is no clean line between "AI music" and "human music" anymore. Auto-tune is AI. Neural amp modeling for guitars is AI. AI-powered mixing boards and mastering software touch nearly every commercial release. A song with human-written lyrics but Suno-generated vocals could be labeled AI on one platform, human on another, and hybrid on a third.
Krawetz surveyed labeling efforts by YouTube (launched 2023), Meta (2024), TikTok (2024), Deezer (2025), and Apple Music (March 2026). The result: complete inconsistency. No two platforms agree on what counts as AI-generated, and none of them can reliably detect it anyway.
This creates what Krawetz calls the "silence as signal" problem. When platforms only label a fraction of AI content, everything without a label looks human-made by default. The absence of a warning becomes a false stamp of authenticity.
Detection Works, But Not at Scale
AI-generated audio does leave artifacts that trained analysts can identify. The problem is that consumers cannot do this analysis themselves, and automated detection systems are fundamentally unreliable in legal or regulatory contexts.
Krawetz points to a specific technical challenge: deep learning-based detectors are non-deterministic, meaning they can produce different results on the same input. That fails the standards courts require for authenticated evidence (FRE Rule 901, Daubert). And the systems need continuous retraining that companies typically neglect after launch.
Even legitimate hardware trips these detectors. The Google Pixel 10's HDR processing merges multiple frames into one image, eliminating the single-frame noise patterns that AI-tampering detectors look for. A real photo from a real camera gets flagged as manipulated.
The Actual Fix Is Curation, Not Classification
Krawetz's most provocative argument: the industry is solving the wrong problem. Instead of asking "is this AI?" platforms should be asking "who made this and will they stand behind it?"
The music industry historically acted as a quality filter. Record labels, radio programmers, and music critics curated what reached listeners. Streaming removed those gatekeepers and flooded every platform with unfiltered content. AI just accelerated a problem that already existed.
His proposed solution is "artist verification" - establishing identity and accountability rather than trying to classify content by production method. A human who uses AI tools to produce great music should not be penalized. A bot farm flooding Spotify with AI-generated ambient tracks should be stopped - but because it is spam, not because it is AI.
This framing matters beyond music. The same four problems Krawetz identifies - detection, labeling, disclosure, and curation - apply to AI-generated text, images, and video. And right now, every platform is repeating the music industry's mistakes: shipping unreliable detection, applying inconsistent labels, and hoping the problem sorts itself out.
It will not.