Subvert.fm has officially published its AI policy, drawing a sharp line in the sand for music platforms navigating the generative AI boom. The cooperative platform explicitly prohibits works where generative AI systems like Suno, Udio, or Boomy serve as the source of creative authorship. However, in a move that will reassure many working producers, the policy permits the use of AI-assisted production tools such as iZotope Ozone, LANDR mastering, and Logic Pro’s AI features. This distinction aims to protect human creative labor while acknowledging that machine learning is already embedded in standard production workflows.

The Core Distinction: Generation vs. Assistance

The policy’s foundation rests on a precise technical definition that many other platforms have failed to articulate. Subvert defines "Generative AI" as systems designed to produce new creative content from prompts, such as text-to-music or text-to-image models. In contrast, "AI-assisted production tools" are defined as software that uses machine learning to analyze, adjust, or enhance existing content without originating it. This means pitch correction, noise reduction, and smart EQ are considered standard technology, not violations. Artists using these tools do not need to disclose their usage, though they are encouraged to list them in optional credits.

Visual Art and Voice Cloning Restrictions

The ban extends strictly to visual media as well. Album art, promotional images, and music videos generated by tools like Midjourney, DALL-E, or Sora are prohibited. There is no grandfathering clause for visual assets; artists uploading existing catalogs must replace AI-generated cover art with human-made alternatives. Regarding audio, AI voice cloning is banned unless the model was trained exclusively on the artist's own recordings or explicit consent was obtained from the person whose voice is being modeled. This protects artists from unauthorized digital likeness exploitation.

Community-Driven Governance and Appeals

This policy wasn’t handed down from on high; it emerged from a lengthy participatory process. Between May 2025 and March 2026, 68 members contributed 352 posts to a forum thread dedicated to crafting the policy. To handle edge cases, Subvert has established a Cooperative Moderation Panel that reviews appeals. Exceptions may be granted for artists with disabilities who use AI as an accessibility accommodation, or for those using self-trained models based on their own original work. The panel will publish anonymized summaries of decisions to build a transparent body of precedent.

Key Takeaways

  • Generative AI (Suno, Udio, Midjourney) is banned for both audio and visual content.
  • AI production tools (LANDR, iZotope, smart EQ) are permitted without mandatory disclosure.
  • Visual assets must be human-made; no grandfathering for existing AI cover art.
  • Appeals are possible for accessibility needs or self-trained models with clear documentation.
  • Subvert commits to not using hosted content to train third-party AI models.

The Bottom Line

Subvert is betting that clarity beats ambiguity. By separating technical ML assistance from generative authorship, they’ve created a policy that is actually enforceable, unlike the vague hand-waving seen elsewhere in the industry.