The explosive rise of generative artificial intelligence has disrupted creative and technical industries, raising an urgent legal question for developers, artists, and enterprises: can AI generated code and artwork be copyrighted under US law? As platforms like Midjourney, ChatGPT, and GitHub Copilot become deeply integrated into professional workflows, the intellectual property framework protecting these outputs is undergoing a trial by fire. Under current United States copyright standards, purely machine-generated outputs cannot be copyrighted. However, the legal reality is nuanced when human creativity merges with algorithmic execution.
1. The Bedrock of US Copyright: The Human Authorship Requirement
To understand the legal boundaries governing generative artificial intelligence, one must look directly at the foundational statutes and precedents of American intellectual property law. The central inquiry of our analysis—can AI generated code and artwork be copyrighted under US law—is met with a definitive roadblock in Section 102(a) of the Copyright Act. This statute states that copyright protection subsists solely in 'original works of authorship fixed in any tangible medium of expression.' Historically, the term 'author' has been interpreted exclusively as a human being.
This human-centric interpretation is backed by a long lineage of federal judicial precedents. In the seminal Supreme Court case Feist Publications, Inc. v. Rural Telephone Service Co. (1991), the Court ruled that copyright protection requires at least a 'modicum of creativity' originating from intellectual human labor. More recently, the famous 'monkey selfie' dispute (Naruto v. Slater, 9th Cir. 2018) solidified the doctrine that non-humans lack standing to claim copyright protection. The court explicitly concluded that the Copyright Act does not extend to works generated by non-humans, a rule that seamlessly applies to machine learning models today.
In March 2023, the United States Copyright Office (USCO) published formal Policy Guidance on Works Containing Material Generated by Artificial Intelligence. The USCO reiterated that it will not register works where the traditional elements of authorship—such as selection, design, execution, and creative expression—are generated entirely by machine algorithms. The core question during registration is whether the work is fundamentally one of human authorship, with the computer or program merely being an assisting instrument.
2. Can AI-Generated Artwork Be Copyrighted under US Law?
Visual art has served as the primary battleground for testing the boundaries of AI copyright. Over the last few years, several high-profile registration rejections and federal court rulings have established a clear boundary: purely prompt-generated art is ineligible for copyright protection.
The Stephen Thaler Case and the 'Creativity Machine'
In the landmark case Thaler v. Perlmutter (D.D.C. 2023), computer scientist Stephen Thaler challenged the USCO's refusal to register a digital artwork titled 'A Recent Entrance to Paradise.' The image had been generated autonomously by Thaler's AI program, the 'Creativity Machine.' Thaler listed the AI as the author and attempted to claim copyright ownership as the owner of the machine. The U.S. District Court for the District of Columbia upheld the USCO's rejection, stating that human authorship is a 'bedrock requirement of copyright' that cannot be bypassed, regardless of how advanced the tool is.
Kristina Kashtanova and the 'Zarya of the Dawn' Decision
Another pivotal administrative decision involved the graphic novel Zarya of the Dawn by creator Kristina Kashtanova. Kashtanova authored the text and compiled the visual layout, but generated the individual background and character illustrations using Midjourney. While the USCO initially granted a full registration, it subsequently revised its decision upon learning of the AI usage. The USCO ruled that while Kashtanova owned the copyright to the text and the compilation (the selection and arrangement of the images), the individual Midjourney-generated images were not protectable. The USCO reasoned that because Midjourney users cannot predict or fully control the specific artistic output of their text prompts, they do not act as the 'mastermind' or creative director of the visual elements.
The Myth of 'Prompt Engineering' as Creative Expression
Many digital creators argue that spending hours crafting complex, multi-layered prompts constitutes sufficient human intellectual effort to justify copyright. However, the USCO rejects this argument. From a legal standpoint, writing a prompt is akin to giving general artistic instructions to a human commission painter. The prompter merely suggests a conceptual direction; the AI model performs the actual creative execution (colors, shadows, line work, composition). Because the AI acts as the primary 'expressive agent,' the resulting image lacks the requisite human authorship.
3. Can AI-Generated Software Code Be Copyrighted?
Under US copyright law, computer software code (both human-readable source code and machine-executable object code) is treated as a literary work. Consequently, the exact same legal standards apply to automated coding as they do to fine art. If developers rely on AI tools like GitHub Copilot, Amazon Q, or Claude to write code, they must tread carefully to avoid losing proprietary control over their software assets.
The Limits of Unmodified AI Code Blocks
If a developer uses generative AI to write an entire function, class, or application module, that specific code block cannot be copyrighted. The logic, structural pattern, syntax, and execution flow are determined by machine neural networks trained on public open-source libraries, not human intellect. If an enterprise includes large blocks of unaltered AI-generated code in their software, those specific sections automatically enter the public domain. This presents a massive vulnerability, as competitors can legally copy and use those public domain code segments without fear of infringement.
The Integration of Human and AI-Generated Software Code
In professional software environments, developers rarely copy-paste generative code directly without modification. Instead, they debug, refactor, and integrate AI-generated suggestions into a larger, human-architected codebase. The original human-written scaffolding, specific system architecture, custom API integrations, and manual logic adjustments are fully eligible for copyright protection. The key challenge lies in proving which sections of the code were human-authored versus AI-generated, and whether the AI-generated portions are 'de minimis' (so minor that they do not impact the overall copyrightability of the file).
4. The Legal Spectrum of Human-AI Collaboration
The refusal of the USCO to protect purely AI-generated outputs does not mean that any work utilizing AI is barred from copyright. Rather, US law seeks to isolate the human creative contribution and protect those specific elements as derivative works or compilations.
The Compilation Doctrine (Section 103)
Under Section 103 of the Copyright Act, compilations of pre-existing materials can be protected. If a human creator selects, coordinates, and arranges AI-generated assets (images, text, or code) in an original and creative manner, the overall layout and sequence can be copyrighted. For example, while a set of individual AI-generated game textures cannot be protected, a human developer's custom arrangement of those textures into a highly unique user interface is copyrightable as a compilation. Crucially, this does not grant the creator exclusive rights to the underlying, AI-generated components themselves.
Substantial Modification and Derivative Works
If a human artist or developer takes an AI-generated output as a base and substantially modifies it, the resulting product may qualify for registration as a derivative work. To meet this threshold, the human's modifications must be 'sufficiently original' to constitute an independent work of authorship. Minor tweaks, standard formatting, basic color correction, or minor code refactoring are typically considered 'de minimis' and will not suffice. However, deep manual digital painting over an AI generation, or rewriting major computational logic within an AI-suggested software function, will qualify the human-modified portions for protection.
5. IP Protection Comparison Matrix
The following table outlines how different scenarios of AI usage in both software development and visual arts are treated under current US copyright standards.
| Asset Type | Purely AI-Generated (Unprotectable) | Human-AI Collaborative (Protectable) |
|---|---|---|
| Visual Artwork | Images generated solely from a text prompt via Midjourney or DALL-E. | AI-generated assets that are heavily edited, manual digital painting added, or compiled into a human-written graphic novel layout. |
| Software Code | Functions, modules, or entire scripts generated natively by ChatGPT or GitHub Copilot without human edits. | Proprietary software architectures combining custom human code integrated with heavily modified, refactored, and restructured AI modules. |
| Written Content | Blog posts, copy, or technical manuals produced entirely by LLMs with basic prompt instructions. | Articles heavily edited, restructured, researched, and supplemented with human-authored commentary and unique creative voice. |
6. Strategic Best Practices for Creators and Enterprises
Navigating the intersection of copyright law and artificial intelligence requires proactive risk management. Failing to establish a clear audit trail of human authorship can jeopardize your entire intellectual property portfolio.
"The US Copyright Office does not require you to completely abandon AI. It requires you to be candid. Developers and artists must meticulously document their creative contributions and disclose the use of generative AI in their applications, or risk having their registrations invalidated post-hoc during litigation."
— Isabella Thorne, Senior IP Counsel at LegalGlobe
To secure your hybrid creations and protect your business assets, follow these fundamental strategic rules:
- Maintain Clear Version Control: Use platforms like GitHub, GitLab, or localized file-history tracking to document the evolution of your code or artwork. Keep logs of the step-by-step human modifications applied to any AI-generated foundations.
- Be Transparent with the USCO: When filing for a copyright registration, you are legally obligated to disclose the use of generative AI if it constitutes more than a 'de minimis' portion of the work. You must disclaim the AI-generated parts and explicitly seek protection only for the human-authored components.
- Implement Internal AI Governance: Companies should establish clear internal guidelines outlining when, where, and how developers can use AI coding tools. Restrict AI usage for core, high-value proprietary modules, and keep a log of AI-assisted versus human-authored source files.
7. The Future of AI Intellectual Property in US Courts
The intersection of artificial intelligence and IP law remains highly volatile. Beyond registration issues, major class-action lawsuits are pending against generative AI developers (such as the high-profile litigation against Stability AI, Midjourney, and Microsoft/GitHub Copilot) alleging that training these models on copyrighted data without consent constitutes mass copyright infringement. The defense of 'fair use' under Section 107 will be a central pillar of these legal battles.
Additionally, Congress is actively debating potential legislative reforms. Some industry advocacy groups argue for a new tier of intellectual property rights specifically tailored to machine-generated creations, while others advocate for maintaining the strict human-centric focus of traditional copyright. Until federal statutes are explicitly rewritten, the 'human authorship' standard stands as the definitive barrier protecting human creators and structuring the modern digital marketplace.