AI Copyright: Rules, Risks, and What Your Business Actually Needs to Do About It

TL;DR

• The U.S. Supreme Court denied certiorari in Thaler v. Perlmutter on March 2, 2026, locking in the rule that human authorship is a bedrock requirement of U.S. copyright law. Fully AI-generated work is unprotectable. Period.
• AI copyright lawsuits crossed into triple digits by mid-2026, with the Copyright Alliance now tracking more than 60 active federal cases. Courts are reaching opposite conclusions on identical legal questions within the same district.
• The biggest practical risk isn’t getting sued for using AI. It’s producing content you can’t legally protect. If a competitor copies your AI-generated asset and you can’t prove sufficient human creative contribution, you have zero legal recourse.
• The EU AI Act’s high-risk provisions take full effect August 2, 2026, and the European Parliament is pushing a mandatory licensing regime that would apply to any GenAI model available in the EU — regardless of where training occurred. The tools you use will change whether you sell in Europe or not.

---

Our content team got a wake-up call last quarter. A competitor scraped three of our highest-performing articles and republished them almost verbatim. When I contacted our attorney, she asked a question that stopped me cold: “Can you prove a human made the creative and expressive decisions in these pieces?”

I couldn’t. We’d been prompting, taking the AI draft, spot-editing, and hitting publish. That workflow meant exactly zero copyright protection.

That question sent me on a deep dive that’s consumed months of my professional life. Here’s the honest version of what I learned — not the law-firm white paper, but the practical answer to what you can and can’t do with AI right now.

The One Rule That Now Governs Everything

Copyright requires a human author. After March 2, 2026, this rule is about as settled as anything gets in IP law.

The Supreme Court refused to hear Thaler v. Perlmutter, leaving the D.C. Circuit’s ruling intact: a work created entirely by AI without human creative input cannot receive copyright protection. Dr. Stephen Thaler had been fighting since 2018 to register an image produced autonomously by his “Creativity Machine.” Every court that reviewed his claim answered the same way.

But here’s what didn’t get resolved. Thaler explicitly disclaimed any human creative input — his argument was that the AI was the true author. The courts never had to answer the harder question: how much human involvement is enough?

That question is alive in Allen v. Perlmutter, pending in Colorado. An artist used 600-plus iterative Midjourney prompts to create an image and is challenging the Copyright Office’s denial of registration. Whatever that court decides will directly shape how much documentation you need to keep. [1]

“The Copyright Act requires copyrightable works to be authored by a human being. AI is a tool, not an author.”

— Baker Donelson, analysis of the March 2, 2026 Supreme Court certiorari denial

The Copyright Office’s January 2025 Part 2 report confirmed that protection exists when “a human author has determined sufficient expressive elements.” Its Part 3 report, released May 2025, went further, finding that fair use does not always shield AI developers from licensing obligations when training on copyrighted materials. [2]

Think of it like a camera. Nikon doesn’t own your photographs. You made the creative choices. Same principle here — but only if you can prove it.

The Lawsuit Landscape in May 2026

Norton Rose Fulbright’s comprehensive March 2026 case analysis identified six landmark rulings. The Copyright Alliance tracker shows new filings arriving monthly. Here’s where the biggest cases stand:

Case	Status	Why It Matters
Bartz v. Anthropic	Settled $1.5B, Sept 2025	~$3K per book across ~482K pirated works. Training = fair use; storing pirated copies = not. Largest copyright settlement in U.S. history. [1]
Kadrey v. Meta	Partial dismissal; “seeding” claims active	Court split from Bartz: ruled training fair use regardless of source legitimacy. Different judge, same district, opposite conclusion on key legal point. [1]
Thomson Reuters v. Ross	TR won; on appeal in 3rd Circuit	First ruling that using copyrighted materials for AI training is NOT fair use. Ross is a search tool, not generative AI — that distinction was critical. [1]
In Re OpenAI MDL	12 cases consolidated in SDNY; 108M output logs ordered produced	Testing whether AI outputs that are “substantially similar” to copyrighted works infringe. OpenAI’s motion to dismiss survived. [1]
Disney + Universal v. Midjourney	Filed June 2025; post-mediation Aug 2026	First major studio lawsuit targeting AI outputs, not just training. Seeks injunction + $150K per willful work. [1]

The most telling move of the past year? Disney’s two-track strategy. It sued Midjourney for infringement in June 2025. Then in December 2025, it signed a $1 billion licensing deal with OpenAI covering 200-plus characters for use in Sora. Litigation sets boundaries. Licensing builds markets. Both are happening simultaneously.

The court split between Bartz and Kadrey is the development to watch. Two judges in the Northern District of California reviewed near-identical facts — AI training on copyrighted books — and reached opposite conclusions on whether the data source matters for fair use. As Norton Rose Fulbright noted, this divergence may push appellate courts to weigh in, but none have ruled yet.

The Risk Nobody’s Talking About

Everyone asks: “Can I get sued for using AI content?”

That’s the wrong question for most businesses. The more common and damaging risk: you build a content library, a competitor copies it, and you have zero standing to stop them. Because you never established sufficient human creative contribution.

The Copyright Office has been clear since 2023 that prompts alone don’t qualify as sufficient human control. Prompt, light edit, publish = you own nothing.

Then there’s the flip side. The In Re OpenAI multidistrict litigation is testing whether AI outputs that are “substantially similar” to copyrighted works constitute infringement. The court found in October 2025 that “some outputs that a reasonable jury could find are substantially similar to plaintiffs’ works.” It then ordered production of 108 million output logs for analysis. [1]

So you face risk on both ends. Your AI content might not be protectable. The tools you use might produce infringing outputs. And most platform terms of service quietly shift liability to you, the end user.

Pro Tip: Enterprise tools like Adobe Firefly and Microsoft Copilot offer IP indemnification for commercial outputs. Free-tier tools almost never do. That single difference decides whether an infringement claim becomes your problem or your vendor’s. Read the ToS.

The AI Copyright Risk Matrix

Here’s the practical framework we now use for every AI-assisted piece:

Content Scenario	Human Involvement	Protection Likelihood	Infringement Risk	Action
AI writes complete draft, you publish as-is	None/minimal	Very low	Moderate	Never do this for anything you need to own
AI generates draft, you substantially rewrite	High	Strong	Low	Save before/after versions; log creative choices
AI generates images with generic prompts	None	Very low	Low-moderate	Fine for internal/social; risky for brand assets
AI generates images mimicking a recognizable style/character	None	Very low	HIGH	Avoid. Disney v. Midjourney is the cautionary tale
AI assists with research/outline, human writes from scratch	High	Strong	Low	Best practice for protectable content

The principle: the more human creative decision-making you can prove, the more defensible your position on both fronts.

Three things to do starting today:

1. Save your drafts. Keep the raw AI output alongside every edited version.
2. Log your creative choices. A sentence of internal notes — “restructured sections 2-4, rewrote intro with original data, cut 40% of AI draft” — creates a defensible paper trail.
3. Check your tool’s terms and IP indemnification. Enterprise plans from Adobe and Microsoft offer contractual protection. Free-tier tools don’t. For high-value content, the premium is cheap insurance. [3]

Europe Is Coming for Your AI Stack

The EU AI Act’s high-risk system provisions take full force August 2, 2026, mandating copyright disclosure, compliance documentation, and risk management systems. But the bigger story is what comes next.

The European Parliament is pushing for a statutory licensing regime that would require GenAI providers to obtain licenses for all copyrighted works used in training and supply an itemized list of every work. The critical provision: rules would apply to any model “placed on the EU market, regardless of where the training occurred.” Non-compliant providers “should be barred from operating within the Union.” [4]

For a marketer in Chicago, this matters because the tools you use will change. If OpenAI or Midjourney must restructure to comply with EU rules, those changes cascade to every user in every jurisdiction.

California adds its own layer. The California AI Transparency Act amendments, signed October 2025 and effective August 2, 2026, require provenance data disclosure in AI-generated content distributed in the state. If your content reaches California audiences, it applies. [5]

How to Protect Your Content Right Now

1. Treat AI as a research assistant, not a ghostwriter. Use it for outlines, data gathering, and brainstorming. Write the final version yourself. The Copyright Office confirms that AI-assisted creation doesn’t bar copyrightability when the human makes the expressive choices. [2]

2. Document everything worth protecting. Save raw AI outputs, version histories, and edit logs. Baker Donelson’s analysis emphasizes that maintaining “records of prompts, iterative revisions, curation decisions, and post-generation creative choices” is now critical practice. [6]

3. Run originality checks before publishing. AI doesn’t “know” when it’s reproducing training data. It predicts the next token. Plagiarism scans catch the obvious stuff — and the obvious stuff is what litigators screen for first.

4. Choose enterprise-tier tools for high-value work. The free version of your AI tool almost certainly shifts all copyright risk to you. Enterprise plans from Adobe Firefly and Microsoft Copilot include IP indemnification. For production-level content, the premium is negligible compared to the exposure. [3]

5. Build an internal AI use policy. Specify approved tools, permissible use cases, and documentation requirements. This protects you if an employee unintentionally generates an infringing output and creates a repeatable workflow that’s easier to defend than ad hoc prompting.

Frequently Asked Questions

Can you copyright AI-generated content in 2026?

Only the human-authored portions. Purely AI-generated work remains unprotectable after the Supreme Court’s March 2, 2026 certiorari denial in Thaler v. Perlmutter. Substantial human editing, rewriting, or creative selection and arrangement of AI outputs creates protectable content — but the human contribution must be meaningful and ideally documented. [2]

What was the Bartz v. Anthropic settlement?

In September 2025, Anthropic paid $1.5 billion to settle a class action by nearly 500,000 authors. The court had ruled training AI on copyrighted books is fair use, but storing pirated copies is not. The ~$3,000-per-book settlement covered past training data. It remains the largest copyright settlement in U.S. history. [1]

Who is liable when AI produces infringing content?

Unresolved as of May 2026. Most platform terms place liability on the end user, but Disney v. Midjourney and In Re OpenAI are testing whether developers share responsibility when their tools produce infringing outputs. Norton Rose Fulbright calls this tripartite question “one of the most complex unresolved issues in AI copyright law.” [1]

Should I stop using AI for content creation?

No. The Copyright Office confirms AI-assisted creation is copyrightable when the human makes sufficient expressive choices. Use AI as a drafting and research tool. Write the final version yourself. Document your creative decisions. That workflow is well-positioned under current law.

---

Building workflows that protect your content isn’t complicated. Document the human craft. Save the receipts. Treat AI output as raw material, not finished product.

If you want help building a legally defensible AI content program, LoudScale works with brands to set up the right tools, workflows, and documentation habits.

Read more: [internal anchor: AI content workflow tools] and [internal anchor: content marketing legal checklist].

Sources

[1] Norton Rose Fulbright, “AI in Litigation Series: An Update on AI Copyright Cases in 2026,” March 2026. https://www.nortonrosefulbright.com/en/knowledge/publications/ce8eaa5f/ai-in-litigation-series-an-update-on-ai-copyright-cases-in-2026

[2] U.S. Copyright Office, “Copyright and Artificial Intelligence Part 2: Copyrightability,” Jan 2025; Part 3: “Generative AI Training” (Pre-publication), May 2025. https://www.copyright.gov/ai/

[3] Microsoft, “Copilot Copyright Commitment,” Sept 2023. https://blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copyright-commitment-ai-legal-concerns/ ; Adobe, “Firefly for Enterprise — Legal FAQs,” updated 2024. https://business.adobe.com/products/firefly-business/firefly-ai-approach.html

[4] Wolters Kluwer, “European Parliament Study Recommends Statutory Licensing as the Optimal Copyright Framework for AI Training,” March 2026. https://legalblogs.wolterskluwer.com/copyright-blog/european-parliament-study-recommends-statutory-licensing-as-the-optimal-copyright-framework-for-ai-training/

[5] California Legislative Information, “SB 1000: California AI Transparency Act,” March 2026. https://calmatters.digitaldemocracy.org/bills/ca_202520260sb1000

[6] Baker Donelson, “Supreme Court Denies Certiorari in Thaler v. Perlmutter,” March 5, 2026. https://www.bakerdonelson.com/supreme-court-denies-certiorari-in-thaler-v-perlmutter-ai-cannot-be-an-author-under-the-copyright-act