Artificial Intelligence Policy
Status: pre-launch draft
Last updated: 25 July 2026
1. Human creativity first
Underground is intended to support human artists and authentic music communities. Generative technology must not be used to appropriate artists’ works, identities or audiences.
2. No training on User Content
Underground will not use User Content to train foundation models, generative AI systems or machine-learning models that generate music, audio, images, text or audiovisual works, and will not license User Content to third parties for those purposes, without specific, informed and explicit prior consent from every affected rightsholder.
Consent for ordinary Platform operation, search, moderation or technical processing is not consent for generative-model training.
3. Prohibited third-party use
Users and third parties may not scrape, download, aggregate or otherwise use Platform content or metadata to build training datasets, embeddings, voice models, style models or generative systems without permission from Underground and all relevant rightsholders.
4. AI-generated and synthetic content
The following are prohibited:
- music or audio generated wholly or substantially by generative AI and presented as an authentic human artist release;
- unauthorised cloning or simulation of a person’s voice, likeness or distinctive identity;
- deceptive impersonation of artists, bands or rights holders;
- AI output that infringes copyright, trademark, privacy, publicity or contractual rights;
- synthetic engagement, reviews, comments or profiles used to manipulate discovery.
Assistive uses such as editing, noise reduction, mastering assistance, accessibility, translation, brainstorming or limited generative elements may be permitted when the resulting work remains meaningfully directed and authored by humans, all inputs and outputs are lawfully used, and any required disclosure is made.
Before launch, the product team must define whether and where AI-assistance disclosure is mandatory.
5. Platform automation
Underground may use machine learning or automated processing for security, abuse detection, moderation support, search, recommendations, accessibility, transcription, metadata assistance and other requested product functions.
Such processing must be documented in the processing register, minimised, subject to vendor and transfer review, and separated from prohibited general-purpose training.
6. Enforcement and review
Suspected synthetic or impersonating content may be restricted while evidence is assessed. Decisions should consider provenance, project files, creator explanation, metadata, detection limitations and other available evidence. Automated detectors alone should not be treated as infallible proof.
Violations may result in removal, upload restrictions, suspension or termination under the Content Moderation Policy.