Legal Notice

This page explains the legal posture behind the BananaRat Studio platform — what we do, what we don't do, and why your songs are sanitized through a cleanroom layer before audio generation.

Your songs belong to you

BananaRat Studio takes a non-exclusive platform license to your compositions for the limited purpose of running the service. We do not claim ownership. We do not take work-for-hire. We do not assign rights to ourselves. The full terms are in our Terms of Service.

Why we sanitize artist references (cleanroom layer)

Every prompt and lyric that reaches our neural audio engine passes through a multi-layer cleanroom filter. If your lyrics or composition prompt mention a real artist by name (band names, vocalist names, or named song titles), the cleanroom replaces those tokens with [artist-redacted] before generation runs.

This is intentional. Three reasons:

  1. Style-imitation legal exposure. Generating audio "in the style of" a named artist creates documented intent that could surface in a copyright dispute. Removing the named reference protects you and us from claims of derivative-work generation.
  2. Right-of-publicity protection. US states (notably Tennessee under the ELVIS Act, California under AB 2602) and most EU jurisdictions recognize a person's right to control commercial use of their name and likeness. Generating an "Anthrax-style" track on demand is a use we don't want to enable on your behalf.
  3. Trainer-side compliance signaling. Combined with the embedded CAWG Training-Data-Mining opt-out in every generated file, the cleanroom forms a coherent "we don't reproduce or impersonate named artists" posture that holds up to scrutiny.

Common false positives we've seen: band names that overlap with everyday words (Anthrax / Nirvana / Madonna / Heart / Queen / Train / Boston / Chicago). The cleanroom uses a 2-tier blocklist (strict + ambiguous) and only redacts ambiguous tokens when paired with style-signal phrases ("in the style of", "sounds like", "à la"). Your "song about a queen bee" is safe; "song in the style of Queen" is not.

When the cleanroom redacts something in your song, you'll see a notice above the player card. The notice expands to show exactly which tokens in which fields were affected.

AuthorMark and C2PA content credentials

Every song you produce carries an embedded C2PA content credentials manifest signed with our cryptographic key. The manifest records:

  • What was generated (prompt summary, no model-name leaks)
  • When (signed timestamp)
  • Your editorial decisions (a tamper-evident hash chain of the choices you made)
  • The mandatory "Contains AI-generated content" disclosure (CA SB 942 + EU AI Act Article 50)
  • A CAWG Training-Data-Mining v1.1 opt-out (machine-readable signal to AI training pipelines that this work declines training use)

You can verify any manifest by uploading the file to contentcredentials.org/verify (Adobe's public verifier). During beta, the trust state may read "Invalid" because we sign with a self-rooted dev cert chain — the cryptographic signature itself is valid; only the trust path isn't externally vouched yet. Once our DigiCert Content Authenticity certificate lands, the trust state will resolve to "Valid" with no manifest changes required.

Honest framing — what AuthorMark is and isn't

What it is: evidence-grade documentation of authorship that reads cleanly into PRO and USCO registration paperwork. The combination of our notation-accurate audio (registration-authoritative) + the C2PA manifest + the Witness hash chain forms a single coherent provenance package.

What it isn't: a copyright claim. We do not register your songs with the US Copyright Office or any PRO on your behalf. We do not guarantee that any song produced by the platform will pass a similarity-screening review or qualify for protection in any jurisdiction. AuthorMark is the evidence layer, not the legal layer.

Registration and royalty eligibility depend on jurisdiction and the registrar's decision. The provenance pack is evidence-grade, not legal advice.

EU AI Act + CA SB 942 compliance posture

Both the EU AI Act (Article 50) and California SB 942 require AI-generated content to carry machine-readable disclosure starting August 2, 2026. BananaRat Studio meets both requirements via:

  • A C2PA manifest in every mastered audio file (component files — stems, MIDI, MusicXML and AAF — do not carry one; those formats either cannot, or are production parts rather than the finished recording)
  • (Q2 2026) An inaudible neural watermark in the mastered audio, which survives re-encoding, loudness normalisation and sample-rate conversion

Other legal pages

Questions? Email feedback@bananarat.com. We will respond.