Copyright & Legal in Generative Media

Generative media sits at the intersection of the most active legal frontiers in AI. Who owns AI-generated content? Can models be trained on copyrighted material? What happens when AI generates content that resembles a real person or a protected work? These questions are actively being litigated, and the answers will shape the industry. This page covers the current state - which is actively evolving.

Training Data: The Core Dispute

Generative models are trained on massive datasets of copyrighted content: photographs, artwork, videos, music recordings, books. The central legal question: is this training permitted under fair use (US) or similar doctrines in other jurisdictions?

Two camps:

• Training is fair use - AI companies argue that training is transformative use: the model does not store or reproduce training data, it learns statistical patterns. Like a human artist studying other artists, training is a form of learning, not copying.
• Training requires licensing - rights holders argue that their works are used without consent or compensation, creating economic harm. The model can generate content that competes with the original creators.

Major Legal Cases (2024–2026)

• Getty Images v. Stability AI - Getty sued Stability AI for training on Getty's image library without a license. The case is proceeding in US and UK courts; Stability AI disputes that training infringes.
• RIAA v. Suno / Udio - The Recording Industry Association of America sued both music generation companies in 2024, alleging their models were trained on copyrighted sound recordings without a license. Settlement discussions are ongoing.
• Authors Guild v. OpenAI - Authors including John Grisham and George R.R. Martin sued OpenAI for training on their books. The case raises questions about whether LLMs can reproduce copyrighted text.
• Midjourney / Stability AI / DeviantArt lawsuit - Three artists sued multiple image generation companies for training on their artwork. Class action certified in 2024.

Who Owns AI-Generated Content?

The US Copyright Office (USCO) has taken a consistent position: AI-generated content alone is not copyrightable. Copyright requires human authorship. Content that is generated by AI without meaningful human creative input cannot be owned by anyone - it enters the public domain immediately.

However, the analysis is nuanced:

• Prompt alone → not copyrightable - A text prompt that generates an image does not constitute sufficient creative expression to warrant copyright in the output.
• Significant human curation and arrangement → may be copyrightable - If a human selects, arranges, and modifies AI outputs in a way that reflects human creative choices, the resulting work (not the AI-generated elements) may be protectable.
• AI as tool → output may be copyrightable - If AI is used as a tool under significant human direction (similar to using Photoshop), the human's creative choices may be sufficient for copyright in the resulting work.

The practical implication: commercially valuable AI-generated media (brand campaigns, stock content) should involve documented human creative decision-making to support any copyright claims.

Platform Terms and Output Licensing

Commercial use rights for AI-generated content vary by platform:

Likeness and Voice Rights

Beyond copyright, two additional rights complicate AI media generation:

• Right of publicity - individuals have rights over commercial use of their likeness, voice, and name. AI-generated videos that realistically depict real people can violate these rights - even without using actual footage of that person. Several US states have passed laws specifically addressing AI-generated likenesses.
• Voice cloning - ElevenLabs, PlayHT, and similar tools can clone a voice from a few minutes of audio. Using someone's cloned voice commercially without consent is both legally and ethically fraught. The FTC has issued guidance; SAG-AFTRA and other unions have negotiated AI voice protections into contracts.
• Deep fakes - several jurisdictions now criminalize non-consensual intimate deep fakes. The EU AI Act classifies systems that generate realistic fake representations of real people as high-risk.

Practical Guidance for Builders

• Use platforms with clear commercial licensing if you plan to commercialize outputs
• Do not generate realistic likenesses of real, identifiable people for commercial use without their consent
• Avoid generating content that stylistically mimics a specific living artist's work in commercial contexts
• Document human creative decisions in the workflow - this supports future copyright claims
• Monitor the legal landscape - this area is changing rapidly; court rulings in 2025–2027 will significantly clarify the picture