AI-Generated Content and Intellectual Property: Who Owns What in 2026

A client walked into our office last month with 200 product images generated by Midjourney. She wanted to register copyrights on all of them. I had to tell her that most of them probably couldn’t be registered. The ones where she’d made significant creative choices during and after generation had a chance. The rest, where she’d typed a prompt and accepted the output, almost certainly did not.

That conversation happens every week now. Businesses are producing enormous volumes of AI-generated content, including marketing copy, product images, code, music, and video, and assuming they own all of it. The legal reality is far less clear. Copyright law in the United States requires human authorship, and purely AI-generated works don’t qualify. The line between “AI-generated” and “AI-assisted” is where the real questions are, and where the money is.

I’m Evan Dotta, a partner at Mister Wolf P.C. I work with technology companies, creative businesses, and individual creators on intellectual property protection. Here’s what the law actually says about AI-generated content ownership in 2026, what recent cases have decided, and what you should be doing right now to protect your work.

Can You Copyright AI-Generated Content?

The Human Authorship Requirement

Copyright protection in the United States requires human authorship. That principle has been the law since the Supreme Court decided Burrow-Giles Lithographic Co. v. Sarony in 1884, when the Court held that a photograph was copyrightable because the photographer made creative decisions about posing, lighting, and composition. The camera was a tool. The human was the author.

The U.S. Copyright Office has applied this principle consistently. Its March 2023 guidance (Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence) made it clear: AI-generated content without human authorship cannot be registered. In January 2025, the Office’s Part 2 AI report reaffirmed that generative AI outputs can be protected only when a human author determines sufficient expressive elements, such as human-authored material perceptible in the output or creative selection, arrangement, or modification. Mere prompts are not enough.

The federal courts agree. In Thaler v. Perlmutter, the U.S. District Court for the District of Columbia ruled in 2023 that a visual artwork autonomously created by an AI system could not be copyrighted when the AI was listed as the author. The D.C. Circuit affirmed in March 2025, holding that the Copyright Act requires all authors to be human beings.

So the baseline is clear: if an AI system generates content without meaningful human creative input, don’t assume you own an enforceable copyright in the output. The Copyright Office may refuse registration, and your competitors may argue the output is free to copy.

The Gray Area: AI-Assisted Works

Pure AI generation sits at one end. Pure human creation sits at the other. Most real-world AI use falls somewhere in between, and that middle ground is where copyright protection may apply.

The Copyright Office’s decision on Zarya of the Dawn (2023) is the most instructive example. Kris Kashtanova created a graphic novel using Midjourney to generate images and then arranged, selected, and sequenced those images with human-authored text. The Copyright Office granted copyright registration for the overall compilation (the human selection and arrangement) and the text, but denied copyright for the individual AI-generated images themselves.

The lesson: human creative choices can be copyrightable even when the raw materials are AI-generated. Selection, arrangement, modification, and editorial judgment all count. Typing a prompt and accepting whatever the AI produces does not.

The Copyright Office’s AI initiative drew more than 10,000 public comments by December 2023, and the Office has since issued a multi-part AI report covering digital replicas, copyrightability, and generative AI training. That process matters because registration examiners are applying the same human-authorship framework to real applications every day.

If you generate content with AI tools, document your creative process. Save your prompts, revision history, selection decisions, and manual edits to AI output. This documentation proves human authorship if you need to register a copyright or defend one in court.

How Are Training Data Lawsuits Changing the Rules?

The Fair Use Battle

The biggest IP fight in the AI industry isn’t about who owns AI output. It’s about who owns the training data that goes in.

Large language models and image generators are trained on massive datasets that include copyrighted works, scraped from the internet, licensed from data brokers, or both. Copyright owners are suing, arguing that using their works to train AI models constitutes copyright infringement. AI companies respond that training is fair use under 17 U.S.C. Section 107.

The most prominent case is The New York Times v. Microsoft and OpenAI, filed in the U.S. District Court for the Southern District of New York in December 2023. The Times alleges that OpenAI copied millions of copyrighted articles to train GPT models and that the models can reproduce Times content nearly verbatim. OpenAI argues fair use. As of May 2026, the case had not produced a final merits ruling, but the stakes are enormous. If training on copyrighted content is not fair use, the economics of generative AI may shift toward licensed data.

Thomson Reuters v. Ross Intelligence, decided on summary judgment in the District of Delaware in 2025, offered an early data point. The court found that Ross Intelligence infringed Westlaw headnotes and rejected the fair use defense, emphasizing the commercial and competitive nature of Ross’s legal research product.

These cases haven’t produced a definitive rule yet. The fair use analysis under Section 107 has four factors: purpose and character of use, nature of the copyrighted work, amount used, and market effect. Courts are reaching different conclusions. But the trend is clear: using copyrighted content to train commercial AI products without a license carries real legal risk.

What This Means for Your Business

If you use AI tools in your business, you’re probably not training models. But you’re using tools built on training data that may include copyrighted material. If a court rules that certain AI models were trained on infringing data, it could affect the legal status of their outputs.

This isn’t a reason to stop using AI tools. The risk sits primarily with the AI companies, not their end users. But it’s a reason to diversify your content strategy. Don’t rely entirely on AI-generated material for business-critical assets like brand identity, flagship marketing campaigns, or core product features.

What Did California’s 2024 AI Laws Change?

AB 2602 and AB 1836

California was the first state to pass targeted legislation addressing AI’s impact on intellectual property rights in the entertainment industry. Two bills signed in 2024, AB 2602 and AB 1836, address the use of AI-generated digital replicas of real people.

AB 2602 added Labor Code Section 927. It restricts the enforceability of contract provisions that allow creating digital replicas of an individual’s voice or likeness in place of personal or professional services the person otherwise would have performed. The protection turns on whether the intended uses are reasonably specific and whether the person had legal counsel or union representation when negotiating the digital replica term. This protects actors, musicians, voice artists, influencers, and other service providers from buried AI rights in standard agreements they sign without focused legal review.

AB 1836 amended California Civil Code Section 3344.1 (the post-mortem right of publicity statute) to explicitly cover AI-generated digital replicas of deceased individuals. Before this, there was ambiguity about whether creating an AI-generated performance of a dead celebrity violated existing law. AB 1836 removes that ambiguity.

These laws primarily affect entertainment, but the principles extend further. Any business using AI to generate content that incorporates a real person’s voice, likeness, or identity needs to consider whether California law applies. See our post on right of publicity in California for more. Given that California is where most entertainment, technology, and digital media companies operate, it usually does.

What Are the Risks of AI-Generated Code and Software?

Open Source Contamination

AI code generation tools like GitHub Copilot, Amazon CodeWhisperer, and similar products are trained on publicly available code repositories. Many of those repositories contain code licensed under open source licenses (GPL, MIT, Apache 2.0, and others) that impose conditions on use, modification, and redistribution.

If an AI tool generates code that substantially reproduces GPL-licensed source code, using it in a proprietary product could trigger the GPL’s copyleft requirements, potentially forcing you to open-source your entire codebase. This isn’t theoretical. The class action lawsuit filed against GitHub, Microsoft, and OpenAI in the Northern District of California in 2022 alleges exactly this kind of harm.

A software startup I worked with discovered during pre-acquisition due diligence that roughly 15% of their codebase came from Copilot. The acquirer’s legal team flagged it as a risk and demanded indemnification for any open source license violations in AI-generated code. The deal closed, but the indemnification cut into the founders’ net proceeds.

Practical steps for code

If your engineering team uses AI code generation tools, implement a policy now. Require developers to review AI-generated code for similarity to known open source projects. Use software composition analysis (SCA) tools to scan for license-restricted code. Document which portions of your codebase were AI-generated and which were human-written. This documentation matters during due diligence, licensing disputes, and copyright registration.

Review your company’s AI tool usage policies this week. If you don’t have a policy, create one. It should cover: who can use AI tools, what output needs human review before use, how AI-generated content is labeled and tracked, and who owns the output under your vendor agreements.

How Should Businesses Protect IP When Using AI Tools?

Review your vendor agreements

Every AI tool has terms of service that address IP ownership of outputs. These terms vary dramatically.

OpenAI’s terms (as of May 2026) assign the user’s right, title, and interest in output to the user to the extent permitted by law, with conditions and the practical warning that multiple users may receive similar or identical outputs. Midjourney’s terms differ. Adobe Firefly’s differ again. Google’s Gemini has specific provisions about enterprise versus consumer use.

Read the terms carefully. If your business depends on owning the content you create with AI tools, confirm that the vendor agreement actually gives you those rights. Some agreements include broad licenses back to the AI company, allowing them to use your inputs and outputs for model training. If your inputs include confidential business information or trade secrets, you have a problem.

Document human creative input

The most important thing you can do to protect the copyrightability of AI-assisted work is document human creative input at every stage:

• Save your prompts (with dates)
• Record which outputs you chose and which you rejected
• Document edits, modifications, and additions to AI output
• Keep version histories showing evolution from AI output to final product
• Note the creative rationale for your decisions

This creates a paper trail showing that a human author made the creative choices that give the work its expressive character. Without it, you’re relying on memory and argument.

Establish clear AI use policies

At Mister Wolf P.C., we advise companies to create written AI use policies covering:

1. Approved tools: which AI tools employees can use for work product
2. Prohibited inputs: trade secrets, confidential client information, and proprietary data cannot enter AI tools
3. Output review requirements: human review and modification of AI-generated content before publication or client delivery
4. Ownership documentation: standard procedures for documenting human creative input
5. Disclosure obligations: whether and when to disclose AI use to clients, customers, or regulators

The companies that build these systems now will have a significant advantage when the law catches up. And it’s catching up. The Copyright Office has proposed new rulemaking on AI and copyright. Congress has introduced multiple bills on AI-generated content. The Central District of California, home to both entertainment and a massive tech sector, will likely decide many of these questions. If you need help, find a Los Angeles intellectual property lawyer who stays current on AI developments.

What Happens Next in AI and Intellectual Property?

The legislative pipeline

Congress has introduced multiple bills on AI and intellectual property since 2023. The most significant proposals would require disclosure of copyrighted training data, create a federal right of publicity, establish licensing frameworks for AI training data, and impose labeling requirements for AI-generated content. The NO FAKES Act of 2025, for example, would create a federal right against unauthorized digital replicas of a person’s voice or visual likeness.

California has already moved faster than Congress in several areas. AB 2602 and AB 1836 took effect in 2025, and 2025-26 proposals such as the Digital Dignity Act show that Sacramento is still looking at digital replica liability, deepfake misuse, and AI-generated identity harms.

The practical reality: the rules are still being written. That creates risk and opportunity. Companies that establish strong IP practices around AI use now will be better positioned than those scrambling to comply with new regulations after they pass.

If you’re producing AI-generated or AI-assisted content and have questions about ownership, registration, or risk, bring your situation to an IP attorney who understands the technology and the current legal framework. The answers depend on the facts: what tools you used, what creative decisions you made, and how you’ll use the output commercially. Gather your AI use records, vendor agreements, and work examples, then schedule a consultation to map out a protection strategy before the law shifts.