The editorial argues Tidal is the first major DSP to publish a parseable, declarative position on AI training, generated music, and artist consent — answering questions Spotify, Apple Music, YouTube Music, and Amazon Music have ducked for 18 months. The 201-point HN ranking is cited as evidence that the clarity itself is novel and valued.
By submitting the policy page directly to Hacker News where it reached 201 points and 231 comments, the submitter signaled that this corporate document was worth surfacing — an implicit endorsement that Tidal's written stance stands apart from competitors' silence on AI training and synthetic content.
The editorial highlights approvingly that Tidal made no fair-use carve-out and no 'transformative use' hedge, treating the entire catalog and metadata as off-limits regardless of who is scraping. It frames this as a clean inversion of the default that labs like Suno and Udio relied on, with opt-in licensing as the only legitimate path.
The editorial calls the disclosure requirement the 'operationally interesting' part of the policy because it pushes flagging responsibility upstream to DistroKid, TuneCore, and other distributors rather than leaving Tidal to detect synthetic uploads after the fact. Undisclosed AI tracks become a contract violation eligible for takedown, which is a concrete enforcement lever competitors lack.
The editorial notes that Tidal's opt-in pathway for audited AI use cases is a quiet but important acknowledgment that some artists actively want their voice or style licensed for AI products. Rather than blanket prohibition, the platform leaves rightsholder agency intact, which the editorial frames as a more mature stance than pure refusal.
Tidal — the Block-owned (formerly Square) high-fidelity streaming service — published a formal AI Policy at tidal.com/ai-policy. The page hit 201 on Hacker News, which is a non-trivial signal for a corporate policy document: most land with a thud. This one didn't, because it answers questions Spotify, Apple Music, YouTube Music, and Amazon Music have spent the last 18 months refusing to answer in writing.
The policy stakes out three positions. First: Tidal's catalog, metadata, lyrics, artwork, and user interaction data are explicitly off-limits for training generative AI models, regardless of whether the scraper is a research lab, a startup, or a hyperscaler. No fair-use carve-out, no "transformative use" hedge. Second: AI-assisted and AI-generated tracks are *permitted* on the platform, but distributors are required to flag them at upload time, and undisclosed synthetic uploads are treated as a content-policy violation eligible for takedown. Third: rightsholders retain the ability to opt their works into specific, audited AI use cases — which is a quiet acknowledgment that some artists *want* their voice or style licensed, and the platform won't block that.
The document is short, declarative, and notably free of the "we believe in responsible AI" boilerplate that makes most policy pages unreadable. That's probably why it ranked.
The streaming industry has been operating under a tacit don't-ask-don't-tell regime since late-2023 ghost-artist scandals and the Drake/Weeknd "Heart on My Sleeve" deepfake. Labels sued Suno and Udio. Spotify quietly removed tens of thousands of AI tracks but published no rule. Apple's terms gesture at "misuse" without defining it. Tidal is the first major DSP to write down, in language a developer or A&R lawyer can actually parse, what its position is.
The disclosure requirement is the operationally interesting part. It pushes the labeling burden onto distributors — DistroKid, TuneCore, CD Baby, Amuse, Stem — which means metadata schemas have to grow a field. Expect a `ai_generated: boolean` and probably a `ai_assistance_level` enum (none / arrangement / vocal-clone / fully-generated) propagating through the ingestion APIs over the next two quarters. That's a real engineering ask, not a press release.
The training prohibition is the legally interesting part. Tidal can't actually stop a determined scraper from grabbing 30-second previews and album art — the audio fingerprinting cat is out of that bag. What the policy does is create a clean cause of action: a model trained on Tidal data is now provably in breach of stated terms, which strengthens any future copyright or contract claim and removes the "we didn't know it was off-limits" defense Suno and Udio have leaned on. It's pre-litigation hygiene dressed up as a policy page.
The community reaction on HN split predictably. The top comment praised the clarity ("finally, an actual sentence instead of a values statement"). The most-upvoted skeptical reply pointed out that Block also owns Cash App and TIDAL has had cash-burn issues — a policy that's easy to publish when you have 5 million subscribers is harder to maintain when a label demands you license the catalog for training in exchange for a renewal. That's a fair point. Policies survive contact with revenue pressure or they don't.
Worth noting: the policy says nothing about Tidal *itself* training models on user behavior — recommendation, mood detection, lyric search. That's a deliberate omission. The fight here is over third-party generative training, not internal ML.
If you ship anything that ingests music metadata — a recommendation system, a tagging service, a music intelligence API, an embedding pipeline — three things just changed. One: assume `ai_generated` will become a required field across major distributor APIs within 6-12 months. Start designing for it now; retrofitting nullable booleans into a schema after the fact is the kind of work that eats sprints. Two: if you're scraping streaming services for training data, you now have explicit written notice from at least one DSP. Your legal exposure went up overnight, even if you never touch Tidal directly — courts read industry-standard policy pages as evidence of what's reasonable. Three: if you're building tools for artists or labels, an "AI provenance" feature — sign-this-track-as-human, or sign-this-as-AI-assisted with C2PA-style attestation — just became a credible product wedge.
For anyone building generative audio models: the era of "we trained on publicly available data" as a legal shield is closing fast, and Tidal's policy is one more brick in that wall. Expect Spotify to publish something within two quarters — either a similar policy or a deliberately weaker one that signals openness to licensing deals. Apple will be last and least specific, because Apple is always last and least specific on policy.
For platform engineers in adjacent industries — Substack, Medium, Patreon, Bandcamp, even Letterboxd — Tidal's document is a usable template. Three paragraphs, no marketing voice, machine-readable intent. Steal the format.
The interesting test isn't whether other streamers follow — they will, in some watered-down form. The test is what happens the first time a major label asks Tidal to carve out an exception so the label's own AI partner can train on the catalog. That's when we'll find out whether "opt-in only" means what the policy says or whether it means "opt-in for everyone except the parties writing the checks." Policies are easy. Enforcement under revenue pressure is the part that actually scores points.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.