Glitch detected. Source traced.
The glitch: 90,000 AI-generated tracks. Every day. Uploaded to Deezer. A single platform. The source: generative models trained on scraped, unlicensed audio. This isn't a prediction. It's a streaming queue.
Crypto Briefing broke the number. The article framed it as a content ownership problem. Fine. But they missed the deeper flaw. The infrastructure of trust is broken. And blockchain, the obvious patch, is nowhere in the conversation.
Context: The Platform as Oracle
Deezer isn't just a music player. It's a centralized oracle. It decides what is real, what gets played, what gets paid. 90,000 AI tracks per day means 2.7 million per month. 32.85 million per year. To put that in perspective: the entire Spotify catalog in 2020 was roughly 60 million tracks. At this rate, AI music will match that in under two years.
This isn't a niche problem. It's a system design failure. Platforms rely on metadata, licensing, and manual curation. None of these scale against a firehose of synthetic audio. The detection algorithms are playing catch-up. They're trained on known model outputs. But models evolve. The gap widens.
Liquidity draining. Logic broken.
The 'liquidity' here is attention. Human artists see their streams diluted. Revenue pools are split among millions of fabricated songs. The logic of the royalty system โ pay per stream to the rights holder โ assumes a finite, attributable set of creators. That assumption is now invalid.
Core: Technical Decomposition
Let's unpack the 90,000 number. It's an upload rate, not a quality metric. Most of these tracks are likely low-effort, bulk-generated noise. Background music for sleep playlists. Lo-fi beats for study sessions. The AI tools are cheap and abundant. Open-source models like AudioCraft, MusicGen. Commercial APIs from Suno and Udio.
But here's the technical catch: Detecting AI audio is an arms race. Every detection model has a recall value. False positives hurt human artists. False negatives let AI flood the gate. Based on my experience audit-smart contracts for NFT metadata, I've seen this pattern before. Off-chain data is a centralization point. You can't verify what you can't see. Here, the 'off-chain' reality is the generation pipeline. The platform doesn't have access to the model weights or training data. It only sees the output WAV.
In 2021, I reverse-engineered the Bored Ape Yacht Club contract and found the metadata could be changed by the dev team without on-chain consensus. Same principle: control the database, control the truth. Deezer controls the detection. Who audits the auditor?
NFT metadata mismatch found.
The mismatch: The NFT community spends millions on on-chain provenance. Music streaming spends nothing. There is no timestamped, immutable record of creation for audio files. A track could be AI-generated, uploaded in 2019, and claimed as human. No one would know. The metadata fields โ 'artist', 'composer', 'label' โ are self-reported. The entire trust model is built on honesty. In a world with incentive to cheat, that's a bug.
Contrarian: The Unreported Angle โ Centralized Gatekeeping
Everyone is focusing on AI vs. artists. The real story? It's AI vs. platforms. Deezer reports these numbers to justify tighter control. They want to be the arbiter of authenticity. They want to decide which songs are 'real' and which are 'synthetic'. That power is dangerous.
Imagine a scenario: A human artist uses AI as a tool โ beat generation, vocal harmonization โ but the detector flags 20% of the track as synthetic. Deezer labels it 'AI-assisted', which penalizes its royalty rate. The artist loses revenue. The platform wins control.
This is the contrarian blind spot. The crypto solution โ on-chain creation proofs, verifiable signatures โ is systematically ignored because it conflicts with the platform's business model. If every track carries a cryptographic hash linking it to a creator wallet, Deezer can't arbitrarily classify. They lose their gatekeeper role.
But the crypto industry remains silent. They're busy shilling meme coins and governance tokens. They forgot their original purpose: trustless verification. A decentralized file system for audio provenance. A protocol where each upload records the generation parameters, model version, and timestamp on-chain. That's not science fiction. That's a weekend project for a skilled Solidity dev.
Takeaway: Watch the Legal Battlefront
The next signal won't be a blockchain partnership. It will be a lawsuit. A human artist will sue Deezer for mislabeling their work as AI. Or a label will sue an AI company for model training without consent. Once the courts force platforms to prove authenticity, the demand for on-chain solutions will spike.
Until then, the glitch remains. 90,000 tracks a day. No provenance. No transparency. The platforms are building walls. The crypto builders are asleep. The code is waiting to be written.