Generative AI Copyright: Who Owns AI-Generated Content?

When you generate an image with Midjourney, draft marketing copy with ChatGPT, or create code with Claude, who owns the result? Can you copyright that AI-generated logo? What if your AI-generated content infringes someone else’s copyright?

These questions sit at the intersection of cutting-edge technology and century-old copyright doctrine—and the answers are more complex than most AI users realize. While AI platforms’ Terms of Service may grant you ownership of outputs, U.S. copyright law tells a different story: purely AI-generated works cannot be copyrighted at all.

Who Owns AI-Generated Content? The Short Answer

Under U.S. law, no one owns a copyright in purely AI-generated content: copyright protects only works of human authorship, a rule the D.C. Circuit affirmed in Thaler v. Perlmutter and the Supreme Court left standing in March 2026. Platform Terms of Service—OpenAI’s, Anthropic’s, and Midjourney’s—may contractually assign you ownership of outputs, but a contract cannot grant copyright protection that federal law denies. Copyright is available only for the human-authored contributions: substantial modification of AI outputs, human-created elements, and the creative selection, coordination, and arrangement of AI-generated material.

This article examines the current legal framework for AI-generated content ownership, analyzes how major AI platforms handle output rights, explores infringement liability risks, and provides practical strategies for protecting AI-assisted creative work.

The Human Authorship Requirement: Copyright Law’s Fundamental Limitation

U.S. copyright law contains a bedrock principle that creates immediate challenges for AI-generated content: copyright protection extends only to works created by human beings.

The Copyright Office’s Position

On March 16, 2023, the U.S. Copyright Office issued formal guidance establishing that “copyright can protect only material that is the product of human creativity.”¹ The guidance makes clear that “when an AI technology determines the expressive elements of its output, the generated material is not the product of human authorship.”

This position was reaffirmed and expanded in the Copyright Office’s January 2025 report on copyrightability of AI-generated outputs, which concluded, in substance, that copyright does not extend to purely AI-generated material or material where there is insufficient human control over the expressive elements.²

The practical implication: If you type a prompt into an AI system and it generates content autonomously, that output lacks copyright protection—regardless of how creative, valuable, or commercially successful it may be.

Case Law: Courts Affirm the Human Authorship Rule

Two landmark matters tested whether AI-generated works qualify for copyright protection. Both failed.

Thaler v. Perlmutter (2023-2026)

Dr. Stephen Thaler sought to register a copyright for a picture titled “A Recent Entrance to Paradise,” created autonomously by his AI system called the “Creativity Machine.”³ Thaler listed the AI as the author and himself as the copyright owner.

The Copyright Office rejected the application, and courts at every level agreed:

• District Court (August 18, 2023): Upheld the Copyright Office’s denial, holding that copyright law extends only to works created by human beings.
• Appeals Court (March 18, 2025): The D.C. Circuit affirmed, stating that “the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being.”⁴

The D.C. Circuit denied rehearing en banc on May 12, 2025, and the Supreme Court denied certiorari on March 2, 2026. With that denial, the D.C. Circuit’s human-authorship holding stands as settled law: a work generated autonomously by a machine, with no human author, cannot be registered.

Importantly, the appellate court clarified that “the human authorship requirement does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence. The rule requires only that the author of that work be a human being—the person who created, operated, or used artificial intelligence—and not the machine itself.”

Zarya of the Dawn (2023)

Kris Kashtanova created a graphic novel using Midjourney to generate all images, then combined them with human-authored text.⁵ The Copyright Office made a split decision:

Protected elements:

• The text (written by Kashtanova)
• The selection, coordination, and arrangement of text and images

Unprotected elements:

• The individual AI-generated images themselves

The Copyright Office reasoned that Midjourney users lack sufficient creative control over the final output. Unlike photographers who control composition, lighting, and timing, AI users provide prompts that the system interprets unpredictably. This lack of control over expressive elements means the images themselves cannot be copyrighted.

The “Sufficient Creative Control” Test

The Copyright Office has articulated a standard for when human involvement crosses the threshold into copyrightable authorship: the human must exercise “sufficient creative control over the work’s expressive elements.”⁶

What doesn’t meet this standard:

• Writing prompts alone, no matter how detailed
• Selecting among AI-generated options
• Minor post-generation tweaks

What may meet this standard:

• Substantial human modification of AI outputs
• Using AI as one tool in a larger creative process controlled by humans
• Creative selection, coordination, and arrangement of AI-generated elements

The Copyright Office emphasizes that “prompts alone do not provide sufficient human control to make users of an AI system the authors of the output” because prompts are merely unprotectable ideas, not expressions.

Sorted by fact pattern, the outcomes reached under the human-authorship requirement line up this way.

Platform Terms of Service: Contractual Ownership vs. Copyright Protection

While copyright law denies protection to purely AI-generated works, AI platform Terms of Service often grant users ownership of outputs. Understanding this distinction is critical.

Read the table for what it is: a map of contractual positions. Every row allocates rights between you and the provider, and no row can create copyright protection that federal law withholds.

OpenAI (ChatGPT, DALL-E)

OpenAI’s Terms of Use, updated December 11, 2024, state: “As between you and OpenAI, you retain ownership rights in Input and own the Output. OpenAI assigns to you all their right, title, and interest in Output.”⁷

Key limitations:

1. Non-Unique Outputs: OpenAI warns that “Output may not be unique and other users may receive similar output.”
2. Subject to Applicable Law: The ownership assignment is qualified by “to the extent permitted by applicable law”—meaning it remains subject to copyright law’s human authorship requirement.
3. Service-Specific Usage Policies: Individual OpenAI products and features may impose additional usage restrictions beyond the base Terms of Use.

Training Data Use: OpenAI may use your inputs and outputs to train models unless you opt out through account settings.

Anthropic (Claude)

Anthropic’s approach distinguishes between consumer and commercial users:

Consumer Terms (Free, Pro, Max accounts):

Anthropic assigns users “all of our right, title, and interest—if any—in Outputs,” but the terms historically implied these were for “internal, non-commercial” use. Recent updates have removed explicit non-commercial language, but the spirit remains that consumer accounts are for personal use.⁸

Commercial/API Terms:

Business customers receive explicit commercial use rights: “Customer owns all Outputs.” Anthropic formally disclaims any interest in customer content and may not use inputs or outputs to train models without permission.

Copyright Defense: Anthropic offers to defend commercial customers from copyright infringement claims arising from their use of Claude’s services or outputs, paying for approved settlements or judgments.⁹

Midjourney

Midjourney’s ownership terms vary based on subscription tier:

Paid Subscribers:

• Own the assets they create and can use them commercially
• Large companies (>$1M annual revenue) must subscribe to “Pro” or “Mega” plans to own their assets

Free Trial Users:

• Governed by Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0)
• Can use images for non-commercial purposes only, with attribution to Midjourney

Important limitation: If you upscale images created by others, those images remain owned by the original creators.¹⁰

Stability AI (Stable Diffusion)

Stability AI’s Stable Diffusion model generally allows broad commercial use without royalties under its community license, though large companies above stated revenue thresholds may need an enterprise license.¹¹ The open-weights nature provides greater flexibility than closed platforms.

The Critical Distinction

Platform Terms of Service create contractual ownership rights that govern the relationship between you and the AI provider. These contracts cannot, however, grant copyright protection that federal law denies. You may “own” the output under your service agreement, but that doesn’t mean you can register it for copyright or sue others for infringement if they independently create similar works.

AI-Generated Content and Infringement Risk: Two-Sided Liability

AI-generated content creates infringement liability risks in both directions: the AI’s training data and the AI’s outputs.

Input-Side Risk: Training Data Infringement

The most contentious copyright question in AI is whether training generative AI models on copyrighted works constitutes infringement. Major litigation is testing this issue now.

The Copyright Office’s Position on Training

The Copyright Office’s May 9, 2025 report on AI training (released as a pre-publication version) concluded that “it is not possible to prejudge litigation outcomes” and that “some uses of copyrighted works for generative AI training will qualify as fair use, and some will not.”¹²

Factors favoring fair use:

• Transformative use of copyrighted works for pattern recognition
• Use of guardrails to prevent generating infringing outputs
• Technical necessity of ingesting entire works for effective training

Factors against fair use:

• Knowing use of pirated or illegally accessed training data
• Commercial purpose of AI systems
• Potential market harm to copyright holders

Landmark Cases: Training Data Litigation

New York Times v. OpenAI/Microsoft (2023-present)

The New York Times sued OpenAI and Microsoft in late 2023, claiming they used millions of Times articles without permission to train ChatGPT.¹³ In an April 4, 2025 opinion, the court issued a mixed ruling on the motion to dismiss: it allowed the core copyright theories—direct infringement, contributory infringement, and trademark dilution—to proceed to discovery, while dismissing the DMCA Section 1202 claims and the unfair-competition/hot-news misappropriation claim. The result was not a clean plaintiffs’ win, but it left the central infringement questions live.

The U.S. cases have since been consolidated for pretrial proceedings as In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143 (S.D.N.Y.) (Stein, J.). In a January 2026 order, the court compelled production of roughly 20 million de-identified ChatGPT conversation logs—an early signal of the discovery stakes in these matters.

The litigation could establish precedent for whether unauthorized training on copyrighted news articles constitutes fair use or infringement.

Getty Images v. Stability AI (2023-present)

Getty Images filed suit in 2023, accusing Stability AI of using its photographs without permission to train Stable Diffusion and asserting copyright infringement, database-right infringement, and trademark infringement (Stable Diffusion sometimes generated images with garbled Getty watermarks).¹⁴ Parallel proceedings ran in the United Kingdom and the United States.

The UK proceeding produced a judgment that significantly narrowed the case. In Getty Images (US) Inc. v. Stability AI Ltd., [2025] EWHC 2863 (Ch), handed down November 4, 2025, the High Court of England & Wales addressed only what remained after Getty abandoned its primary claims. Getty had dropped its core training-based copyright-infringement and database-right claims because the evidence did not show that Stable Diffusion was trained or developed in the United Kingdom; as a result, the court did not decide whether AI training on copyrighted images infringes. On the surviving secondary copyright-infringement claim, the court ruled for Stability AI, reasoning that the model weights never stored or reproduced the copyrighted works and so Stable Diffusion was not an “infringing copy.” Getty prevailed only on what the court itself characterized as “extremely limited” trademark findings tied to a small number of watermark examples. The net result was a narrow, mixed outcome that was predominantly a win for Stability AI on the copyright questions.

In the United States, Getty’s case has been refiled in the Northern District of California, where the underlying training and infringement questions remain to be litigated under U.S. law.

Bartz v. Anthropic (2025)

Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal.), produced one of the most closely watched—and most misread—rulings in this area. In an Order on Fair Use dated June 23, 2025, Judge William Alsup issued a split decision.¹⁵ He held that training Claude on lawfully acquired books was fair use, describing the use as “exceedingly” and “quintessentially transformative,” and that digitizing lawfully purchased print copies for that purpose was likewise fair use. But he held that downloading and retaining millions of pirated copies of books to build a permanent central library was infringing, and sent that piracy liability toward trial on damages.

Anthropic subsequently agreed to a settlement valued at approximately $1.5 billion—on the order of $3,000 per work for roughly 500,000 books—to resolve the piracy liability. The court granted preliminary approval on September 25, 2025; final approval remains pending following a fairness hearing held May 14, 2026. The case is therefore best understood not as a clean fair-use victory for AI developers, but as a bifurcated holding: lawful-acquisition training can be fair use, while building a library from pirated sources is not.

A companion ruling pointed the other way on its facts. In Kadrey v. Meta Platforms, No. 3:23-cv-03417 (N.D. Cal.), Judge Vince Chhabria granted summary judgment to Meta on June 25, 2025—but on the narrow ground that the plaintiffs failed to develop an evidentiary record of market dilution, not a broad blessing of AI training. Together, Bartz and Kadrey underscore that fair-use outcomes in this space turn heavily on the specific facts and the record each side builds.

Output-Side Risk: When AI Generates Infringing Content

Even if training is lawful, AI systems may generate outputs that infringe copyrighted works—particularly when prompted to do so.

Who’s liable when AI outputs infringe?

Both the user and the AI company may face liability:

1. User liability: If you prompt an AI to generate content similar to a copyrighted work, you could be directly liable for creating an infringing derivative work.
2. Platform liability: AI companies may face contributory or vicarious infringement liability if they facilitate user infringement.

Example scenarios:

• High risk: “Generate an image in the style of [famous living artist]” or “Write a chapter continuing [copyrighted book series]”
• Lower risk: Generic prompts that don’t target specific copyrighted works
• Platform protection: Most AI platforms implement guardrails to refuse prompts seeking copyrighted content reproductions

Practical implications: Don’t use AI to recreate or closely mimic existing copyrighted works. If an AI output looks suspiciously similar to known copyrighted material, conducting a reverse image search or text comparison before commercial use is prudent.

Work-for-Hire and AI-Generated Content

Traditional work-for-hire principles don’t map neatly onto AI-generated content, creating additional ownership complications.

Standard Work-for-Hire Rules

Under the Copyright Act, works created by employees within the scope of employment belong to the employer, as do certain specially commissioned works covered by written agreements.¹⁶

AI Complicates Work-for-Hire

AI systems cannot be employees or independent contractors because they lack legal personhood and cannot enter binding contracts. This creates several scenarios:

Scenario 1: Employee Uses AI to Create Content

If an employee uses ChatGPT to draft marketing materials within their job scope, the work-for-hire doctrine should apply to any copyrightable elements—meaning the human-created aspects (editing, arrangement, creative direction) belong to the employer, but purely AI-generated portions remain unprotected by copyright.

Scenario 2: Contractor Delivers AI-Generated Work

If you hire a contractor to create content and they use AI extensively without disclosure:

• If the final work contains sufficient human authorship, the contractor initially owns the copyright (absent a written work-for-hire agreement)
• If the work is purely AI-generated, neither party has copyright protection
• Contractually, you may still “own” the work under your service agreement, but without copyright enforcement rights

Best practice: Include AI disclosure requirements in contractor agreements and specify whether AI use is permitted.

Scenario 3: Multiple Contributors (Human + AI)

When multiple humans contribute to a work that also incorporates AI-generated elements, determining copyright ownership requires analyzing:

• Which elements were created by humans with sufficient creative control
• How those elements were selected, coordinated, or arranged
• Whether the human contributions are independently copyrightable

International Approaches: EU and UK Diverge from U.S. Law

Not all jurisdictions share the U.S. requirement of human authorship.

European Union

The EU’s Artificial Intelligence Act (Regulation (EU) 2024/1689), published in the Official Journal on July 12, 2024 and in force from August 1, 2024, requires providers of general-purpose AI models to disclose information about copyrighted training data but does not directly address AI authorship.¹⁷ EU member states have indicated that AI-generated content could be eligible for copyright protection only where the human input in the creative process was significant.

The EU approach emphasizes transparency over categorical prohibition, requiring providers of general-purpose AI models to “draw up and make publicly available a sufficiently detailed summary about the content used for training,” including content protected by copyright. The transparency obligations for general-purpose AI models began applying on August 2, 2025.

United Kingdom

UK law takes a notably different approach. The Copyright, Designs and Patents Act 1988 includes provisions for “computer-generated works”—works generated by computer in circumstances such that there is no human author.¹⁸

For computer-generated works:

• Copyright protection is available
• The “author” is whoever “undertook the arrangements necessary for the creation of the work”
• Duration is 50 years from creation (shorter than the life-plus-70 standard)

In principle, this means UK users may have copyright protection for AI-generated outputs that would be unprotectable in the U.S. That protection should not be treated as settled or permanent, however. Section 9(3) is under active government review: the UK Intellectual Property Office’s “Copyright and AI” consultation expressly asked whether the computer-generated-works provision should be retained, amended, or removed, and government conclusions on the question are pending. Businesses relying on UK protection for AI outputs should monitor the outcome of that review.

Practical Implications for International Operations

If your business operates internationally:

• Content that’s unprotectable in the U.S. may currently receive protection in the UK, subject to the pending review of section 9(3)
• EU transparency obligations may require disclosing AI use even where U.S. law doesn’t
• Licensing agreements should specify which jurisdiction’s law governs IP ownership
• Enforcement strategies may vary by jurisdiction based on local AI copyright rules

Practical Strategies: Protecting AI-Assisted Creative Work

Given the uncertain copyright landscape, how can businesses and creators protect their AI-assisted work?

Strategy 1: Maximize Human Creative Control

Structure your AI workflow to ensure substantial human involvement in expressive elements:

Instead of this:

1. Enter detailed prompt
2. Accept AI output with minimal changes
3. Use immediately

Do this:

1. Generate multiple AI outputs as raw material
2. Substantially edit, modify, and refine outputs
3. Combine AI elements with significant human-created content
4. Make creative decisions about selection, arrangement, and coordination
5. Document your creative process

Why it works: The more human creative control over final expressive elements, the stronger your copyright claim.

Strategy 2: Hybrid Creation Workflows

Use AI as one tool in a larger human-driven creative process:

• Writing: Use AI for initial drafts, then extensively rewrite and restructure
• Design: Generate AI elements, then compose them into original layouts with human-created elements
• Code: Use AI for boilerplate code, then write custom logic and architecture
• Art: Generate textures or references with AI, then create original compositions

The Copyright Office has indicated that works containing AI-generated material may be copyrightable if humans contribute sufficient creative authorship in the “selection, coordination, or arrangement” or through “creative modifications of the outputs.”

Strategy 3: Document Your Process

Maintain records demonstrating human authorship:

• Save iteration history showing your refinements
• Document creative decisions and modifications
• Keep original human-created elements separate from AI-generated components
• Note the specific AI tools used and when

Why this matters: If you later seek copyright registration or face an infringement dispute, documentation proves your level of creative control and human authorship.

Strategy 4: Copyright Registration Disclosure

When registering works containing AI-generated elements with the Copyright Office:

• Disclose AI use honestly in your application
• Identify which portions are AI-generated
• Claim copyright only in human-authored elements and creative arrangements
• Provide explanation of your human creative contributions

Failing to disclose AI use can result in registration denial or cancellation.

Strategy 5: Contractual Protection Where Copyright Fails

Since purely AI-generated works lack copyright protection, use other legal mechanisms:

Trade Secret Protection: Keep AI-generated materials confidential when possible, protecting them through non-disclosure agreements rather than copyright registration.

Trademark Protection: Brand names, logos, and trade dress may receive trademark protection even if created with AI assistance (subject to use in commerce and distinctiveness requirements).

Contract Terms: When licensing or assigning AI-generated content, use detailed contracts specifying ownership, use rights, and restrictions—creating contractual obligations even where copyright doesn’t apply.

Terms of Service: If you provide AI tools or outputs to others, your terms of service define permissible uses and can provide breach of contract remedies.

Strategy 6: Watermarking and Attribution

Emerging technologies help identify AI-generated content, though they’re not foolproof:

Technical Watermarking: Google’s SynthID and similar tools embed invisible watermarks in AI-generated images, audio, and text. While these can be degraded by determined actors, they provide some attribution protection.

Visible Attribution: For images, consider visible watermarks or metadata indicating AI assistance (though this may not be commercially desirable).

C2PA Standards: The Coalition for Content Provenance and Authenticity (C2PA) is developing technical standards for content provenance, allowing creators to embed tamper-evident metadata about how content was created.

EU AI Act Compliance: If you operate in Europe, the AI Act requires labeling of certain AI-generated content, making watermarking and attribution increasingly essential.

Strategy 7: Clearance and Risk Assessment

Before using AI-generated content commercially, assess infringement risks:

1. Reverse Image Search: For AI-generated images, use Google Images, TinEye, or similar tools to check for similar existing works
2. Text Similarity Checking: For written content, use plagiarism detection tools to identify potential infringement
3. Style vs. Expression: Distinguish between unprotectable style inspiration and potentially infringing expression
4. Prompt Design: Avoid prompts explicitly requesting imitation of copyrighted works
5. Platform Guardrails: Choose AI platforms that implement content filtering to reduce infringement risk

Strategy 8: Insurance and Indemnification

Consider risk transfer mechanisms:

Commercial AI Platforms: Anthropic and some other commercial AI providers offer indemnification for copyright claims arising from outputs. Review these protections when selecting platforms.

Media Liability Insurance: Traditional media liability policies may not cover AI-generated content. Discuss AI use with your insurer and consider specialized cyber or technology E&O policies.

Vendor Agreements: When hiring contractors or agencies that use AI, require indemnification for copyright claims and disclosure of AI use.

Looking Forward: The Uncertain Future of AI Copyright

The legal framework for AI-generated content remains in flux, with several key developments to monitor:

Pending Legislation

Congress is considering multiple bills addressing AI and copyright, including proposals to:

• Clarify fair use standards for AI training
• Create compulsory licensing regimes for training data
• Establish copyright protection for AI-assisted works meeting certain standards
• Require disclosure of AI use in commercial content

No comprehensive federal AI copyright legislation has passed as of September 2026, and the Supreme Court’s March 2, 2026 denial of certiorari in Thaler leaves the D.C. Circuit’s human-authorship holding as the controlling appellate statement.⁴ That could change as the technology matures and case law develops.

Ongoing Litigation

The outcomes of major cases—particularly the consolidated In re OpenAI litigation and the U.S. refiling of Getty Images v. Stability AI—will shape how courts apply fair use doctrine to AI training and outputs. These cases may establish precedent for years to come.

Copyright Office Rulemaking

As of September 2026, Part 3 of the Office’s Copyright and Artificial Intelligence report, on generative AI training, remains in the pre-publication form released May 9, 2025.¹² The Office continues to refine its guidance on AI and copyright, with additional reports and policy statements expected. It has indicated it will provide further clarity on:

• What level of human creative control suffices for copyright protection
• How to evaluate copyrightability on a case-by-case basis
• Registration procedures for AI-assisted works

International Harmonization Efforts

As different jurisdictions adopt varying approaches to AI copyright, international treaties and trade agreements may eventually harmonize rules—or codify divergent standards that create compliance challenges for global businesses.

Key Takeaways

On Copyright Protection:

1. Purely AI-generated content is not copyrightable in the U.S. under current law
2. The human authorship requirement focuses on creative control over expressive elements
3. Prompts alone are insufficient; substantial human modification or creative arrangement is required
4. Documentation of human creative contributions strengthens copyright claims

On Platform Ownership: 5. AI platform Terms of Service grant contractual ownership but cannot override copyright law 6. Different platforms impose different restrictions on commercial use and revenue thresholds 7. Commercial/API agreements typically provide broader rights than consumer accounts

On Infringement Liability: 8. Both AI training on copyrighted works and AI outputs may create infringement liability 9. Fair use analysis for training data remains fact-specific and unsettled, with major litigation pending 10. Users who prompt AI to generate infringing content may face direct liability

On Practical Protection: 11. Maximize human creative control throughout the creation process 12. Use hybrid workflows combining AI generation with substantial human authorship 13. Employ contractual protections, trade secrets, and trademarks where copyright is unavailable 14. Conduct clearance searches and risk assessments before commercial use 15. Consider platform indemnification and specialized insurance for AI projects

When to Consult Legal Counsel

Given the evolving and complex nature of AI copyright law, consult experienced intellectual property counsel when:

• Developing AI tools or platforms for commercial distribution
• Creating substantial AI-generated or AI-assisted content for commercial use
• Facing copyright claims related to AI training or outputs
• Drafting agreements for contractors or employees using AI tools
• Operating internationally with AI-generated content
• Seeking copyright registration for works containing AI elements
• Structuring IP ownership in AI-assisted collaborations
• Implementing AI in regulated industries with specific content requirements

---

The intersection of generative AI and copyright law presents unprecedented challenges for creators, businesses, and policymakers. While current law clearly establishes that purely AI-generated works lack copyright protection, the boundaries of “sufficient human authorship” remain contested and fact-specific.

As AI capabilities expand and legal frameworks evolve, the strategic approach is to maximize human creative control, document your contributions, implement risk mitigation measures, and stay informed of legal developments. The companies and creators who thoughtfully navigate this landscape today will be best positioned when the law stabilizes tomorrow.

---

Need AI Copyright Guidance?

Astraea Counsel advises AI companies and content creators on copyright ownership, AI-generated works protection, and intellectual property strategy. We help clients navigate the evolving legal landscape of AI and copyright law. Explore our AI & Emerging Tech services.

Related Resources

• California’s Frontier AI Law: SB 53 Compliance Guide - State AI regulation framework
• Federal AI Regulation Landscape: What’s Coming in 2025-2026 - Pending federal AI legislation
• AI Training Data Copyright: Fair Use, Licensing, and Infringement Risk - Training data copyright compliance
• Vibe Coding and Copyright: Who Owns Your AI-Generated Codebase? - Ownership of AI-generated source code
• AI & Emerging Technology Practice - Comprehensive AI legal counsel
• Contact Us - Discuss your AI copyright needs

---

Disclaimer: This article provides general information only and does not constitute legal advice. Consult qualified legal counsel for advice on your specific situation.

---

Founders and platforms weighing ownership risk in shipped AI features can also work directly with our AI and autonomous agent counsel .

Footnotes

1. U.S. Copyright Office, Copyright Registration Guidance: Works Containing Material Generated by Artificial Intelligence, 88 Fed. Reg. 16,190 (Mar. 16, 2023), available at https://www.federalregister.gov/documents/2023/03/16/2023-05321/copyright-registration-guidance-works-containing-material-generated-by-artificial-intelligence. ↩

2. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (Jan. 29, 2025), available at https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-2-Copyrightability-Report.pdf. ↩

3. Thaler v. Perlmutter, No. 1:22-cv-01564 (D.D.C. Aug. 18, 2023), aff’d, 130 F.4th 1039 (D.C. Cir. Mar. 18, 2025), cert. denied, No. 25-449 (U.S. Mar. 2, 2026), available at https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf. ↩

4. Thaler v. Perlmutter, 130 F.4th 1039 (D.C. Cir. 2025) (rehearing en banc denied May 12, 2025; certiorari denied, No. 25-449 (U.S. Mar. 2, 2026)). ↩ ↩² ↩³

5. U.S. Copyright Office, Re: Second Request for Reconsideration for Refusal to Register Zarya of the Dawn (Correspondence ID 1-5270) (Feb. 21, 2023), available at https://www.copyright.gov/docs/zarya-of-the-dawn.pdf. ↩ ↩²

6. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, supra note 2. ↩ ↩² ↩³ ↩⁴ ↩⁵

7. OpenAI, Terms of Use (effective Dec. 11, 2024), available at https://openai.com/policies/terms-of-use/. ↩

8. Anthropic, Consumer Terms of Service and Commercial Terms of Service, available at https://www.anthropic.com/legal/consumer-terms and https://www.anthropic.com/legal/commercial-terms. ↩

9. Anthropic, Expanded Legal Protections and Improvements to Our API, available at https://www.anthropic.com/news/expanded-legal-protections-api-improvements. ↩

10. Midjourney, Terms of Service, available at https://docs.midjourney.com/hc/en-us/articles/32083055291277-Terms-of-Service. ↩

11. Stability AI, Licensing (Community License and Enterprise License), available at https://stability.ai/license. ↩

12. U.S. Copyright Office, Copyright and Artificial Intelligence, Part 3: Generative AI Training (Pre-Publication Version, May 9, 2025), available at https://www.copyright.gov/ai/Copyright-and-Artificial-Intelligence-Part-3-Generative-AI-Training-Report-Pre-Publication-Version.pdf. ↩ ↩²

13. The New York Times Co. v. Microsoft Corp., No. 1:23-cv-11195 (S.D.N.Y. Apr. 4, 2025) (Opinion and Order on Motions to Dismiss); In re OpenAI, Inc. Copyright Infringement Litigation, No. 1:25-md-03143 (S.D.N.Y.) (Stein, J.) (consolidated U.S. proceedings), docket available at https://www.courtlistener.com/docket/69879510/in-re-openai-inc-copyright-infringement-litigation/. ↩

14. Getty Images (US) Inc. & Ors v. Stability AI Ltd., [2025] EWHC 2863 (Ch) (High Court of Justice, Nov. 4, 2025), available at https://www.judiciary.uk/judgments/getty-images-v-stability-ai/. ↩

15. Bartz v. Anthropic PBC, No. 3:24-cv-05417 (N.D. Cal. June 23, 2025) (Order on Fair Use), available at https://www.courtlistener.com/docket/69058235/bartz-v-anthropic-pbc/. ↩

16. 17 U.S.C. § 101 (definition of “work made for hire”); 17 U.S.C. § 201(b) (work made for hire ownership). ↩

17. Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act), 2024 O.J. (L series, July 12, 2024), available at https://eur-lex.europa.eu/eli/reg/2024/1689/oj. ↩

18. Copyright, Designs and Patents Act 1988, c. 48, §§ 9(3), 12(7) (UK), available at https://www.legislation.gov.uk/ukpga/1988/48/section/9; UK Intellectual Property Office, Copyright and AI consultation (open Dec. 17, 2024; closed Feb. 25, 2025), available at https://www.gov.uk/government/consultations/copyright-and-artificial-intelligence. ↩