HN Debrief

Tidal AI Policy

  • AI
  • Media
  • Copyright
  • Marketplaces
  • Music

Tidal says it will keep accepting AI-generated music, tag tracks it detects as AI, remove impersonation and fraud, and stop paying royalties or allowing direct sales for music it identifies as wholly AI-generated. Its terms define AI content as work wholly or substantially generated by generative AI with limited human input, while also admitting detection will have false positives and false negatives. That set the frame for most reactions. People generally liked the idea of cutting off the money spigot to slow the flood of low-effort uploads, especially because many have already seen fake releases show up under real artist names or inside autoplay and discovery feeds.

If you run any marketplace or media platform, this is a concrete example of using monetization policy rather than outright bans to curb generative spam. The harder problem is not principle but enforcement and incentives, because vague definitions and weak detection can turn a cleanup policy into a recommendation and trust problem.

Discussion mood

Cautiously positive. Most people liked demonetizing and labeling AI music because it targets spam and impersonation without pretending a total ban is enforceable, but they were uneasy about vague definitions, unreliable detection, and the possibility that Tidal could profit by promoting unpaid AI tracks.

Key insights

  1. 01

    Unpaid tracks change platform incentives

    Removing royalties does not just punish upload spam. It also creates a catalog that costs the service less to serve, which means ambient playlists, autoplay, and recommendation slots become financially tempting places to insert AI tracks. That makes the policy better for margin than for artists unless recommendation rules are constrained too. The comparison to Spotify’s existing playlist economics sharpened the point. The risk is not only slop flooding in from outside, but platforms quietly preferring music that carries no payout obligation.

    Watch recommendation surfaces, not just upload rules. If your business pays out per item consumed, any zero-royalty inventory will distort ranking unless you explicitly design against it.

      Attribution:
    • AlexandrB #1
    • mtrovo #1
    • elicash #1
    • k__ #1
  2. 02

    This is really a discovery and curation failure

    The strongest frame was that AI music becomes toxic when search and recommendation are already weak. People tolerate endless generic background tracks until those tracks crowd out finding real artists, spoof artist pages, and make feeds unusable. Several commenters argued that labels, tastemakers, or stores like Bandcamp work better because they provide provenance and curation, not just a giant undifferentiated pool. The flood exposed a product failure that existed before AI and is now harder to ignore.

    If users complain about generative slop, do not treat it as only a content policy issue. Invest in provenance, curation, and recommendation controls, because better filtering is what preserves trust.

      Attribution:
    • datsci_est_2015 #1
    • sdellis #1
    • bko #1
    • bunderbunder #1
    • doug_durham #1
  3. 03

    The hard line is between generation and assistance

    The policy sounds crisp until you apply it to actual production workflows. A generated bassline, AI lyrics, an AI vocal, AI stems inside a human arrangement, or heavily produced electronic tracks all sit in the gray zone between tool and author. Commenters noticed that Tidal’s own terms rely on phrases like “wholly or substantially” and “limited human creative input,” which are broad enough to preserve discretion but too vague to guide creators. That fuzziness is not a side issue. It is the policy.

    If you publish an AI policy, define reviewable thresholds and examples before launch. Otherwise support, appeals, and creator trust become the real product you are shipping.

      Attribution:
    • calny #1
    • fxwin #1
    • mvdtnz #1
    • annagio_ #1
    • 6thbit #1
  4. 04

    Audio detection will break on hybrid music

    Detection optimism ran into working musicians and technically minded producers who pointed out how brittle audio forensics can be. Spectral oddities, sample-heavy workflows, and aggressive digital signal processing already exist in legitimate music, so “unnatural” signatures are not a clean tell. At the same time, spam operators can move from one-shot generation toward AI-controlled digital audio workstation workflows that look much more like conventional production. That means obvious fully generated tracks may be catchable, but the policy gets weaker exactly where enforcement matters most.

    Assume classifiers can catch only the low end of abuse. Build appeals, provenance metadata, and conservative enforcement around that limit instead of promising reliable detection.

      Attribution:
    • butlike #1
    • marmarama #1
    • waffletower #1
    • bob1029 #1
    • totallygeeky #1
  5. 05

    Artist impersonation is the immediate user harm

    The most concrete pain people reported was not abstract concern about machine art. It was fake songs landing under real artist names like Yes, BT, and Rush, or recommendation systems surfacing bogus releases as if they were legitimate catalog updates. That turns AI into a fraud and identity problem before it becomes a taste problem. People were far more aligned on stopping impersonation than on whether AI music should exist at all.

    Prioritize identity verification and catalog integrity before debating broader AI policy. Users forgive experimentation more easily than they forgive being tricked about who made the thing.

      Attribution:
    • gwbas1c #1 #2
    • somehnguy #1
    • yellowapple #1
    • postalcoder #1
  6. 06

    Human-first alternatives already exist

    Bandcamp came up repeatedly as the practical counterexample. People liked that it supports direct payment, downloadable files, simple profiles, and discovery rooted in communities rather than engagement maximization. Subvert was mentioned in the same spirit as a cooperative response centered on artists. The thread’s subtext was that marketplaces structured around ownership and direct support are naturally more resilient to AI spam than all-you-can-eat streaming feeds.

    If your audience cares about human creators, business model matters as much as moderation. Direct commerce and smaller-scale curation reduce the upside of spam far more effectively than tagging alone.

      Attribution:
    • jmuguy #1
    • MisterTea #1
    • Sandbag5802 #1
  7. 07

    Copyright uncertainty is doing policy work here

    A lot of support for demonetization rested on the idea that fully generated works may lack copyright protection, at least in the United States for raw model output. That gives Tidal a plausible legal and moral basis for refusing royalties on wholly generated tracks, even if the platform still benefits from hosting them. But commenters also noted that hybrid works remain murky and jurisdiction-specific, so the legal clarity applies mostly to the easiest case. The neat rule ends where mixed authorship begins.

    Do not build product policy on the assumption that AI copyright law is settled. Use the current legal edge cases to shape obvious enforcement, but expect hybrids to stay contested and operationally messy.

      Attribution:
    • gonzalohm #1
    • thewebguyd #1 #2 #3
    • oasisbob #1

Against the grain

  1. 01

    Allowing any AI still concedes too much

    A smaller but forceful group rejected the compromise entirely. From this view, once Tidal hosts AI music at all, it legitimizes the category, keeps the clutter, and still profits from attention even while claiming to protect artists. Labeling and demonetization do not fix the core problem for listeners who want a human-only service. They only soften it while normalizing the presence of the thing they wanted excluded.

    If your brand promise is creator-first, partial accommodation may satisfy no one. Be clear whether you are optimizing for contamination control or for a hard identity position, because users notice the difference.

      Attribution:
    • paxys #1
    • p-e-w #1
    • cush #1 #2
    • romanovcode #1
  2. 02

    For many listeners the source does not matter

    Several commenters simply do not care whether music is machine-made if it works for the job. Background coding music, focus tracks, gym playlists, and mood matching were treated as utility products, not artist relationships. From that standpoint, Tidal is solving a cultural problem some users do not have. They want cheap, abundant, style-specific audio and are happy if AI supplies it.

    Expect audience segmentation, not one market reaction. Utility listening and fandom listening behave differently, so product controls should let each group choose instead of forcing a single norm.

      Attribution:
    • bko #1
    • rvnx #1
    • echelon #1
    • techterrier #1
  3. 03

    Ranking could beat outright demonetization

    One practical objection was that Tidal may be fighting the wrong layer. If users dislike low-quality or deceptive music, a better ranking and recommendation system could downrank high-skip, low-satisfaction, or fake-engagement tracks regardless of how they were produced. That approach treats AI as one weak signal among many instead of as a special forbidden category. It also avoids trying to settle impossible authorship questions in product policy.

    If you already operate a ranking-heavy marketplace, test whether quality and fraud signals solve more of the problem than source classification alone. AI status may be useful metadata without being the main enforcement primitive.

      Attribution:
    • rvnx #1
    • wredcoll #1

In plain english

AI ↩
Artificial intelligence, software systems that generate or analyze content in ways that mimic tasks usually associated with human intelligence.
Suno ↩
An artificial intelligence music generation tool that creates songs from prompts and was repeatedly referenced as an example of current AI music output.

Reference links

Platform policies and company statements

Detection and AI music references

Artist support and self-hosted listening tools

  • MusicBrainz Picard
    Recommended for organizing and tagging downloaded music collections from stores like Bandcamp.
  • Streaming Sucks blog post
    Suggested as a guide to setting up personal music streaming with Subsonic-compatible tools.

Streaming economics and legal background