The Ruling That Broke the Internet
Yesterday, the Supreme Court issued its decision in Anderson v. Stability AI — and if you’re going by headlines, you’d think it either saved or destroyed AI art depending on which outlet you read.
Here’s what the ruling actually says, what it means for AI artists, and why most coverage is missing the point.
The Case: What Was At Stake
Background: In 2024, artist Sarah Anderson sued Stability AI, Midjourney, and DeviantArt, claiming that training AI models on copyrighted images constituted infringement. She sought:
- Damages for past training ($2.5 billion)
- Injunction against future training on copyrighted works
- Recognition that AI-generated outputs are derivative works
The Question: Does training an AI model on publicly available copyrighted images constitute “fair use” under 17 U.S.C. § 107?
The Decision: 6-3 Majority Opinion
Written by: Chief Justice Roberts
The Holding:
“Training machine learning models on lawfully obtained copies of copyrighted works, for the purpose of extracting statistical patterns and relationships, constitutes fair use when the resulting model does not reproduce or distribute substantially similar copies of the original works.”
Translation: Training is fair use. Outputs are a separate question.
Key Points from the Majority
1. Training ≠ Copying
“The process of analyzing works to extract abstract patterns, without retaining or reproducing the works themselves, does not constitute copying in the copyright sense.”
What this means: AI companies can continue training on copyrighted images without permission or compensation.
2. Outputs Are Separate
“Whether AI-generated outputs infringe copyright depends on their substantial similarity to specific training inputs, not the training process itself.”
What this means: If your AI output looks too much like a specific training image, you can still be sued. The training is fine — the output might not be.
3. Transformative Use
“Machine learning that extracts patterns, styles, and techniques from copyrighted works for the purpose of generating new, non-similar works is transformative under fair use analysis.”
What this means: AI art generation is considered “transformative” — similar to how a human artist learns from studying other artists’ work.
The Dissent: What the Minority Argued
Written by: Justice Sotomayor (joined by Kagan and Jackson)
Key Argument:
“The majority’s decision effectively eviscerates copyright protection for visual artists. When an AI system can ingest millions of copyrighted works and produce outputs that compete directly with the originals, the economic value of copyright is nullified.”
The Concern: The dissent argued that AI outputs do compete with original artists’ work — and allowing uncompensated training destroys the market for human-created art.
What This Means for AI Artists
✅ What’s Now Protected
1. Using AI Tools You can legally use Midjourney, Stable Diffusion, Flux, and other AI art tools without fear of copyright liability for the training itself.
2. Selling AI Art AI-generated outputs are legal to sell — as long as they don’t substantially resemble specific copyrighted works.
3. Training Your Own Models You can train models on publicly available images for personal or commercial use.
⚠️ What’s Still Risky
1. Direct Style Mimicry If you prompt for “in the style of [living artist]” and the output is substantially similar to their work, you can still be sued for output infringement.
2. Training on Pirated Data The ruling specifies “lawfully obtained copies.” Training on pirated datasets is still infringement.
3. Trademark Issues Copyright is separate from trademark. Generating Mickey Mouse, Marvel characters, or branded content is still legally risky.
What This Means for AI Companies
✅ Wins
- Training is legal — No need for licensing deals for training data
- Fair use precedent — Strong legal foundation for ML training
- Innovation protected — Can continue developing models without copyright chill
⚠️ Remaining Risks
- Output liability — Can still be sued if outputs infringe
- State laws — Some states may pass AI-specific regulations
- International divergence — EU, UK, Japan have different rules
The Human Authorship Question (Still Unanswered)
Notably, the Supreme Court did not address whether AI-generated works can be copyrighted themselves. That question remains governed by the Copyright Office’s position:
“Works generated by non-humans without creative input from a human author are not copyrightable.”
What this means:
- Pure AI generation (prompt only) = No copyright protection
- Human-edited AI art = Copyright in human modifications only
- Traditional art = Full copyright protection
Industry Reactions
AI Companies (Pro-Ruling)
Stability AI Statement:
“This decision affirms that machine learning is fundamentally a transformative technology. We remain committed to respecting artists while advancing AI innovation.”
Midjourney:
“Fair use protects both human learning and machine learning. Today’s decision recognizes that principle.”
Artist Groups (Anti-Ruling)
Concept Art Association:
“This ruling legalizes the mass theft of artists’ work. We will continue fighting through legislative channels.”
Illustrators’ Partnership:
“The Supreme Court has chosen technology companies over working artists. Congress must act.”
What Comes Next
Legislative Battles
Expect continued pressure for federal AI legislation:
Proposed Bills (2026):
- NO FAKERY Act — Requires AI content labeling
- ART Act — Requires opt-in consent for training
- AI Transparency Act — Disclosure requirements for AI companies
International Divergence
European Union:
- AI Act requires training data transparency
- Opt-out mechanism for copyright holders
- Stricter than US approach
United Kingdom:
- Currently consulting on AI and copyright
- Leaning toward US-style fair use
- Decision expected Q3 2026
Japan:
- Already has broad AI training exceptions
- Most permissive major jurisdiction
- Attracting AI companies post-ruling
Practical Advice for AI Artists
Do This ✅
- Document your process — Keep prompts, iterations, edits
- Add human creativity — Edit outputs in Photoshop, combine elements
- Avoid direct artist mimicry — Don’t prompt “in the style of [specific artist]”
- Use licensed datasets — When possible, train on properly licensed data
- Register copyrights — For works with significant human input
Don’t Do This ❌
- Don’t sell direct copies — If it looks exactly like a known work, don’t sell it
- Don’t ignore takedowns — If you get a DMCA, respond appropriately
- Don’t claim pure AI as human-made — Be transparent about AI use
- Don’t train on pirated data — Use legitimate sources only
The Bottom Line
The Supreme Court’s ruling is neither the salvation nor the destruction of AI art. It’s a pragmatic middle ground:
- Training is protected — AI development can continue
- Outputs are scrutinized — Infringement claims still possible
- Human authorship matters — Copyright still requires human creativity
For AI artists: The legal landscape is clearer, but you still need to be thoughtful about what you create and how you create it.
For traditional artists: The ruling doesn’t help with AI competition, but it does preserve your right to sue over outputs that infringe your specific works.
For everyone: This isn’t the end of the debate — it’s the beginning of a new phase in the AI copyright wars.
Join the Conversation
What’s your take on the ruling?
- 👍 Fair use protects innovation
- 👎 Artists deserve compensation
- 🤷 It’s complicated — both sides have points
Discuss on X @PlotTwist_Daily or in the comments below.
Arty Craftson is a Media Producer at Potter’s Quill Media, covering AI technology, copyright law, and digital creativity. This analysis is for informational purposes only and does not constitute legal advice.
Related Reading: