Skip to contentOpen Music
Business.
Back to Listening Studio

Season 1 · Episode 6 · Music Creation · ~16 min

Human Credit, Clean Metadata & AI Rights: Protecting Your Catalog Right Now

What U.S. copyright rules say about AI-assisted music, why ISRC/metadata hygiene is your first defense, and practical opt-out steps.

Next Moves & Resources
Next Moves & Resources

Human Credit, Clean Metadata & AI Rights: Protecting Your Catalog Right Now

Recap, Workbench tools, and closing disclaimer.

Script verified — run npm run studio -- voice ep-06-protecting-your-catalog-in-the-ai-era && npm run studio -- mix ep-06-protecting-your-catalog-in-the-ai-era --copy-public to activate video playback.

Chapters (6)

Live Synced Transcript (61 turns — click any line to jump)

Episode takeaways

What you’ll leave with

Maya and Sal cut through hype and panic around AI in music by connecting four Open Music Business guides—tracing U.S. human-authorship thresholds for AI-assisted tracks, clean studio workflow and upload hygiene, three-layer metadata and persistent ISRCs as catalog provenance, and realistic AI opt-out steps.

  • Keep three questions separate for any AI-involved track: what copyright law protects, what your tool's contract permits, and what distributors or streaming platforms require or enforce.
  • Under current U.S. Copyright Office guidance and Thaler v. Perlmutter, purely AI-generated material and prompts alone generally do not establish human authorship, while qualifying human lyrics, melodies, performances, arrangements, and edits may be protected case by case—with disclosure required for more-than-de-minimis AI material.
  • Treat AI mastering as a starting point compared at matched loudness, treat separated stems as editable estimates, never upload confidential or third-party audio without upload rights, and save dated session records and plan terms.
  • Maintain a three-layer catalog record outside your distributor (composition, sound recording, release) and follow the central ISRC rule: one distinct recording gets one persistent 12-character code, with new codes for remixes, edits, live versions, and music videos.

Source guides combined

Read the full manuals

Workbench tools

Read the full episode transcript (61 turns)

Artist: Okay, picture this. You spend three weeks recording forty tracks of real instruments and vocals—live drums, acoustic guitar, harmonies, lyrics you rewrote five times. Except a four-bar synth bed in the bridge started from an AI generator prompt.

Editor: Now you're staring at your distributor upload screen and the copyright form, wondering whether four generated bars just erased protection for the entire song.

Artist: Exactly. Half the internet says AI means zero copyright. The other half says paying for a Pro plan hands you full ownership. Those can't both be true.

Editor: Neither one is the whole picture. Protecting your catalog right now means separating copyright law from software contracts, documenting human authorship, and backing every release with clean metadata and recording codes.

Artist: Welcome to Open Music Business: On the Record. I'm Maya.

Editor: And I'm Sal. One ground rule on every episode: Open Music Business is educational only—not legal, accounting, tax, or royalty advice. These episodes are scripted from our source-backed guides and voiced with custom synthetic studio personas.

Artist: And every track title, collaborator, or studio scenario we walk through today is a fictional worked example built to illustrate how the rules operate—not real artist disputes. Today we're connecting four guides to protect your catalog in the AI era.

Artist: Let's build a fictional worked example to carry through the whole episode. Say producer Jordan and songwriter-vocalist Tessa are finishing a single called Glass Harbor. Jordan prompts an AI tool for ten bridge sketches, picks one four-bar synth texture, and drops it into a session.

Editor: Then the humans take over. In our fictional scenario, Tessa writes the lyrics and vocal melody and sings the lead takes, while Jordan writes the main chords, replaces the bass line, records live guitars and drums, rearranges the structure, and mixes forty human-recorded tracks around those four generated bars.

Artist: So when Jordan and Tessa ask whether Glass Harbor is copyrighted, what is the right order of questions?

Editor: Our AI-Generated Music and Copyright guide separates three distinct questions. First: does copyright law protect any part of the result? Second: what does your contract with the AI service permit? Third: how will distributors and streaming platforms treat the track?

Artist: Start with U.S. copyright law—because people treat 'made with AI' like one big yes-or-no switch.

Editor: That's mistake number one. Under the U.S. Copyright Office's January twenty twenty-five Part Two Copyrightability report, purely AI-generated material—or output lacking sufficient human control over expressive elements—is not protected. And in Thaler v. Perlmutter, the D.C. Circuit affirmed that a work must be authored in the first instance by a human, while expressly recognizing that human-made work assisted by AI may be registrable.

Artist: What if Jordan spent two hours refining a detailed prompt to get that synth bed? Doesn't careful prompting count as authorship?

Editor: Under current generally available technology, the Copyright Office says prompts alone generally do not give the user sufficient control over expressive elements. Effort and curation aren't automatically legal authorship.

Artist: <short pause> Okay, so those untouched four bars aren't protected. What about Tessa's lyrics and vocals, Jordan's melody, bass line, live instruments, and arrangement?

Editor: Those human contributions may qualify case by case—perceptible human-authored lyrics or melodies, human performances, creative modifications, and creative selection, coordination, or arrangement—while excluding the underlying AI-generated passage itself. There is no universal percentage or edit count that guarantees protection; it's always fact-specific.

Artist: And when they file a U.S. copyright registration for Glass Harbor, do they have to mention the AI bridge?

Editor: Yes, if the AI-generated material is more than de minimis. Current U.S. registration guidance requires applicants to disclose appreciable AI-generated material and describe the human contribution; the registration can then cover the human authorship while disclaiming the AI material. Leaving it out puts the claim at risk.

Artist: Now hit questions two and three: paid AI service contracts, and streaming platform rules.

Editor: A software contract can't create copyright where law says none exists. Suno's terms are a clear illustration: paid Pro and Premier users receive an assignment of Suno-owned output generated during the paid subscription, yet Suno disclaims any warranty that copyright will vest. Separately, platforms set their own rules: Spotify describes AI-use disclosure metadata and anti-spam measures, while Deezer removes detected AI tracks from algorithmic recommendations and editorial playlists.

Artist: And outside the United States—same rules?

Editor: Always check the jurisdiction. The UK has Section Nine-Three for computer-generated works, yet a March twenty twenty-six UK government report flags unresolved tension with originality doctrine and separate sound-recording questions. Japan's twenty twenty-four official overview also puts simple-prompt output outside copyright—and highlights a vital rule everywhere: copyrightability and infringement are separate. Even without strong rights in an AI output, similarity to an existing song still creates infringement risk.

Artist: Now bring this back into everyday production. Most musicians aren't prompting whole songs—they're using AI mastering, stem splitters, or MIDI sketch tools. Do those carry the same risks?

Editor: No. Our Using AI in Music Production guide splits them into three distinct lanes: mastering assistants, stem-separation tools, and generative composition services. Each creates a different technical workflow and a different rights question.

Artist: Start with mastering. Say Jordan finishes the Glass Harbor stereo mix and tries an AI mastering tool like LANDR or iZotope Ozone's Master Assistant.

Editor: Those analyze a stereo mix and propose or apply EQ, compression, loudness, and stereo-image adjustments—Ozone explicitly calls Master Assistant a starting preset. Treat AI mastering as a starting point, not a verdict. Always compare against your raw mix at matched loudness across multiple playback systems so extra volume doesn't fool you into flattening the groove, and keep final calls human.

Artist: Next: stem separation. In our fictional session, suppose Tessa wants to lift a vocal harmony from an old stereo demo whose multitrack is lost, using a tool like Moises.

Editor: Remember that stem splitters estimate vocals, drums, bass, and guitar—they don't cleanly recover the original studio tracks. Quality varies across source types and models, and bleed or transient artifacts happen. Solo every stem, audition it in context, and archive the original bounce, separated files, service name, date, and your edits.

Artist: And before Tessa uploads any file to a cloud tool, there's a big trap musicians walk straight into.

Editor: Uploading confidential stems, co-written demos, or third-party recordings by default. Having an audio file on your drive isn't the same as having permission to upload it to an AI vendor. Run our three-question production rights map before uploading or releasing.

Artist: Give me all three questions in order.

Editor: One: do you have the right to upload the source material—covering co-writers, performers, samples, and label-owned masters? Moises' terms require upload rights and state user content isn't used to train models without express opt-in, though that's specific to Moises—never assume every tool matches. Two: does your exact service plan permit commercial release? Generative tools like AIVA separate non-commercial use with attribution, limited monetization, and full-copyright tiers by plan. Save a dated copy of those terms.

Artist: <breath> And question three brings us back to human authorship: what did we actually create, and where is our proof?

Editor: Save dated DAW sessions, lyric drafts, raw vocal and instrument stems, and mix notes. Also steer clear of voice imitation without consent review—voice and likeness rights often turn on contracts, state publicity law, or voice-cloning statutes. And note that AI training on copyrighted music remains unsettled policy and litigation—the Copyright Office's training report was still listed as pre-publication in our cited overview—so don't assume training is categorically licensed or settled.

Artist: Alright, Jordan and Tessa documented their human work on Glass Harbor. Now let's talk release day—because when platforms are flooded with automated tracks and disputed claims, sloppy metadata destroys your paper trail.

Editor: Our Metadata Best Practices guide treats metadata as a connected catalog record, not a midnight upload form. It answers four practical questions: what is this release, which exact recording is on it, who created and controls the underlying song, and where should every credit and payment match?

Artist: Walk through the three-layer model using our fictional track Glass Harbor.

Editor: Layer one is the composition record: official work title, Tessa and Jordan as writers with roles, any publisher or administrator, and ownership or collection shares. Layer two is the sound recording record: main artist, recording title, version, duration, recording owner, producer, performers, and its I-S-R-C. Layer three is the release or product record: the single, EP, or album package—release title, sequenced display artists, label, distributor, release date, cover art, explicit status, genre, track order, and U-P-C or E-A-N product code.

Artist: Why do those three layers have to stay separate instead of living in one flat spreadsheet?

Editor: Because the U.S. Copyright Office, DDEX delivery standards, and royalty registries treat compositions, recordings, and products as distinct objects. Put an album code in a recording field, or assume the performing artist owns the composition, and matching fails. DDEX's Electronic Release Notification rules even separate official titles from subtitles or version labels, and main display artists from additional artist roles.

Artist: And I bet the same three mistakes trip up almost every DIY release—starting with skipping producer and performer credits because the upload form marks them optional.

Editor: Right alongside inconsistent spelling or punctuation between your distributor, your P-R-O, and The M-L-C—and keeping your only clean metadata record trapped inside one distributor dashboard. If you switch distributors or challenge a bad credit later, you need your own authoritative catalog file.

Artist: So how do Jordan and Tessa run a repeatable pre-release metadata audit?

Editor: Maintain one release sheet, one row per recording, and one row per composition from a single source of truth, and confirm names, roles, and shares with collaborators before upload. Search The M-L-C first to prevent duplicate works and attach recording details to aid matching. Save your final metadata export, filenames, identifier list, and confirmations. After release, periodically check The M-L-C and SoundExchange—including SoundExchange's catalog tools and Music Data Exchange—remembering that registration supports matching rather than legally settling ownership disputes by itself.

Artist: Let's zero in on the most misunderstood piece of that recording layer: the I-S-R-C. Artists either think it proves copyright ownership, or they mint a new I-S-R-C every time they re-upload a song.

Editor: Let's ground both in the IFPI I-S-R-C Handbook and ISO thirty-nine-oh-one. An International Standard Recording Code is a persistent twelve-character alphanumeric identifier for one specific sound recording or music video—not the composition, lyrics, or album package. And by itself, an I-S-R-C does not prove current ownership, register copyright, grant a license, or guarantee royalties.

Artist: Quick breakdown of those twelve characters—especially the part people misread.

Editor: Three elements: a five-character prefix code, a two-digit year of reference, and a five-digit designation code, typically shown with hyphens and the I-S-R-C label. The trap is that two-digit year of reference—it records the year the code was assigned, not necessarily when the track was recorded or released. And the prefix code isn't live proof of current ownership either, since masters change hands.

Artist: Let's test it on our fictional track Glass Harbor. Jordan and Tessa release the studio master as a single in twenty twenty-six, put the exact same master on an EP later, include it on an unchanged compilation, and eventually switch distributors. How many I-S-R-Cs?

Editor: Just one. The central rule is one distinct recording, one I-S-R-C—never reuse a code for a different recording. When the master is unchanged, it keeps its original I-S-R-C across new formats, bitrates, releases, compilations, territories, distributor switches, and ownership changes.

Artist: Now suppose they release a shorter radio edit, a guest remix, a live acoustic take, and an official music video using the studio audio. Same code?

Editor: New codes for each materially distinct recording. Remixes, edits, live takes, covers, extended versions, and materially creative remasters generally need new I-S-R-Cs—IFPI notes duration changes over ten seconds in relevant cases can trigger a new code, while routine remastering generally keeps the original. And a separately exploited music video needs its own I-S-R-C separate from the audio master, which platforms like YouTube explicitly require.

Artist: Who can assign them, and what reference data goes alongside each code?

Editor: A recording owner, exclusive licensee, or authorized I-S-R-C manager. You can obtain assignment means through your territorial agency—the R-I-A-A in the U.S.—or the International I-S-R-C Agency where no territorial agency exists, regardless of trade-association membership. Authorized distributors like DistroKid can also generate codes or accept existing ones, though workflows and fees vary by provider. Beside every code, IFPI recommends retaining main artist, track title, version title, duration, content type, and publication date.

Artist: <short pause> Now connect all of this to the question every artist asks: once our files and I-S-R-Cs are clean, can we actually opt our catalog out of AI training?

Editor: Here is the honest truth: there is no universal opt-out button that retroactively removes published music from every AI model, and viral copy-paste social posts aren't valid legal objections. Real protection is layered. Keep unreleased demos, isolated vocals, and confidential stems private. Use real account controls where available—like YouTube's third-party AI training setting, noting it covers third-party access rather than Google's own systems, plus platform data-sharing toggles and regional privacy objection flows. Ask your distributor, label, or publisher in writing whether they license your catalog or metadata for AI training. And keep your copyright registrations, split sheets, session logs, and I-S-R-Cs organized so any copyright complaint or voice-clone report is backed by hard provenance.

Artist: Let's turn all four guides into one weekly protection routine before your next release.

Editor: Step one: document human authorship as you go—save dated DAW sessions, lyric drafts, raw stems, and a brief log of any AI tools and subscription terms used. Step two: sign co-writer shares early with the Split Sheet Generator on openmusicbusiness.com, and map your pre-release audit dates with the Release Timeline Planner so credits and I-S-R-Cs never get rushed.

Artist: Step three: keep your three-layer metadata record outside your distributor, preserve I-S-R-Cs on unchanged masters, and assign new codes for remixes, edits, live takes, and music videos. Step four: check the AI Protection Hub on openmusicbusiness.com for platform-by-platform opt-out status, account privacy steps, and written distributor questions.

Editor: Find links to all four source-backed guides, primary Copyright Office and IFPI references, our Workbench tools, and the AI Protection Hub at openmusicbusiness.com/podcast. As always, this show is educational only—not legal, accounting, or tax advice—so consult a qualified entertainment attorney for specific registration or rights questions. Thanks for listening to Open Music Business: On the Record.

Editorial limits & transparency

Educational only. Open Music Business is not a law firm, rights administrator, or accounting service. Copyrightability, platform terms, AI opt-out controls, and registration rules vary by jurisdiction, facts, and date. Consult a qualified entertainment attorney for individualized legal advice. Scripted from Open Music Business source-backed guides and voiced with custom synthetic studio personas. Read how this site is made · Report a correction