Authored by: Jeremy J. Gustrowsky and Charles P. Schmal

Artificial intelligence has fundamentally changed the way software is written, and tools like Claude Code, Cursor, Devin, Windsurf, Lovable, and GitHub Copilot sit at the center of one of the most consequential legal conversations happening in the technology world today. Developers using these tools can generate functional, sophisticated software at a pace that would have been unimaginable just a decade ago, translating plain language instructions into working code across dozens of programming languages. Anthropic’s Claude Code offers deep terminal integration and agentic workflows, Cursor has become a favored AI-native code editor, Cognition Labs’ Devin markets itself as an autonomous software engineer, Windsurf (formerly Codeium) provides an AI-first IDE experience, and Lovable enables non-technical users to generate entire web applications from prompts. As these tools become embedded in professional software development workflows, they are raising questions about intellectual property author/inventorship (and consequently ownership), copyright liability, and legal responsibility that the existing framework of law is only beginning to grapple with.

The debate centers on a deceptively simple question: Who is the author of AI generated code ? Under current United States copyright law, the answer remains unsettled by the courts, though the Copyright Office has recently sharpened its position considerably. In its January 2025 report, Copyright and Artificial Intelligence, Part 2: Copyrightability, the Office reaffirmed the longstanding principle that “[c]opyright protects the original expression in a work created by a human author, even if the work also includes AI-generated material,” while making clear that “[c]opyright does not extend to purely AI-generated material, or material where there is insufficient human control over the expressive elements.” According to the copyright office, the controlling inquiry is “whether the ‘work’ is basically one of human authorship, with the computer merely being an assisting instrument, or whether the traditional elements of authorship in the work . . . were actually conceived and executed not by man but by a machine.”

For developers using tools like Claude Code, Cursor, or Windsurf, the most consequential aspect of the 2025 report is its treatment of prompts. The Office concluded that “given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.” In the Office’s words, “[p]rompts essentially function as instructions that convey unprotectible ideas,” and “[w]hile highly detailed prompts could contain the user’s desired expressive elements, at present they do not control how the AI system processes them in generating the output.” Repeatedly revising or resubmitting prompts does not solve the problem either, because “[b]y revising and submitting prompts multiple times, the user is ‘re-rolling’ the dice, causing the system to generate more outputs from which to select, but not altering the degree of control over the process.” Tools like Devin and Lovable, which are designed to operate with greater autonomy and less line-by-line human intervention, raise this concern even more acutely, since the developer’s role shifts from writing code to reviewing and accepting AI-produced deliverables.

The Office has identified pathways by which developers can secure copyright protection for AI-assisted output. “Human authors are entitled to copyright in their works of authorship that are perceptible in AI-generated outputs, as well as the creative selection, coordination, or arrangement of material in the outputs, or creative modifications of the outputs.” In practical terms, a developer who writes original code, feeds it into an AI tool for enhancement, and whose original expression remains perceptible in the final output will hold copyright in that perceptible expression. Likewise, a developer who meaningfully edits, restructures, or arranges AI-generated code contributes copyrightable authorship in those modifications and arrangements, even though the underlying AI-generated material itself remains unprotected. Whether these contributions clear the originality bar is, as the Office emphasized, a case-by-case determination.

The question of training data adds another layer of complexity to the IP conversation surrounding these coding tools. Large language models are trained on enormous datasets drawn from publicly available text and code, which inevitably includes open source repositories, developer forums, and code shared across the internet. Critics have argued that this training process may constitute copyright infringement if protected code was used without license or permission, and several lawsuits targeting AI companies on exactly these grounds are currently working their way through the federal court system. The outcomes of cases such as Doe v. GitHub, which targets GitHub Copilot and its underlying OpenAI technology, are being closely watched because they are likely to establish precedent that will apply broadly to all AI coding assistants. Anthropic, OpenAI, Anysphere, and other AI developers have made varying investments in responsible development practices, but the legal questions around training data liability remain open as of this writing.

The concept of derivative works further complicates the IP picture. Copyright law protects not only verbatim copying but also the creation of works that are substantially similar to protected originals. If a tool like Claude Code, Cursor, or Devin generates a function or algorithm that closely mirrors a copyrighted implementation, a court could potentially find that the output constitutes an infringing derivative work, even if the resemblance was entirely unintentional and the developer had no knowledge of the original. This risk is not unique to AI generated code, as human developers face the same issue when they inadvertently reproduce protected code, but the speed and volume at which AI tools generate output makes systematic review and clearance more challenging than it would be in a purely human development context. Autonomous agents like Devin, which can produce large volumes of code with minimal human oversight, magnify this concern considerably.

Patent law introduces other dimensions to consider. While copyright protects the specific expression of code, patents can protect the underlying methods, processes, and systems that software implements. A developer who uses Claude Code, Windsurf, or Lovable to build a novel software product may independently arrive at a solution that is already covered by an existing patent, with no way of knowing this from the generated output alone. Conversely, the question of whether an AI system can be named as an inventor on a patent application has been tested in courts around the world, and the answer has been uniformly negative to date.

The United States Court of Appeals for the Federal Circuit confirmed in 2022 in Thaler v. Vidal that inventors must be human beings under current patent law, meaning that even if these tools contribute what might functionally be described as inventive insight to a solution, the patent rights must be claimed by and attributed to the human developers involved. The USPTO reinforced and clarified its position on this issue in November 2025, when it rescinded its earlier February 2024 inventorship guidance and replaced it with a streamlined framework for AI-assisted inventions. Under the current guidance, AI systems are treated as tools, analogous to laboratory equipment, research databases, or software libraries, and the same conception-based inventorship standard applies whether or not AI was used in the development process. Importantly, the USPTO abandoned the earlier approach that had applied the Pannu joint inventorship factors to evaluate a human’s contribution relative to AI output, recognizing that Pannu addresses joint inventorship among multiple natural persons and has no application when only one human is working with an AI tool.

The practical takeaway for developers using Claude Code, Cursor, Devin, and similar tools is that a patentable invention still requires a human to have formed “a definite and permanent idea of the complete and operative invention” in their own mind, and applications that name an AI system as an inventor, whether alone or jointly, will be rejected. The 2025 guidance also extends to design and plant patents, and it addresses priority claims, so that a U.S. application cannot claim priority to a foreign application naming an AI system as the sole inventor, and any foreign or PCT application listing an AI as a joint inventor must be filed in the United States listing only the natural person inventors.

For businesses building products on top of AI generated code, the practical implications of these unresolved legal questions are significant. Companies should be conducting IP audits of AI generated code with the same rigor they would apply to any other externally sourced software, regardless of whether the code came from Claude Code, Cursor, Devin, Windsurf, Lovable, or another AI assistant. Legal teams should be developing clear policies about how AI coding tools are used, how outputs are reviewed and modified, and how copyright and patent searches are integrated into the development workflow. Contracts with clients and partners should address the question of AI generated content directly, including representations about authorship, ownership and indemnification provisions that account for the possibility of future legal developments. The legal landscape surrounding AI and IP is evolving rapidly, and organizations that treat these questions as someone else’s problem today may find themselves unprepared for the consequences tomorrow.

It is also worth recognizing that these AI coding tools are not inherently adversarial to the IP system. Used thoughtfully and with appropriate legal oversight, they have proven potential to democratize software development in ways that could benefit inventors, entrepreneurs, and developers who lack the resources to build large engineering teams. Lovable and similar low-code AI platforms are already enabling founders without traditional engineering backgrounds to build functional products, and tools like Cursor and Windsurf are dramatically accelerating what small development teams can accomplish. The challenge for lawmakers, courts, and the legal profession is to develop frameworks that protect the rights of original creators while allowing the benefits of AI assisted development to be realized broadly. The Copyright Office has taken the position that “[q]uestions of copyrightability and AI can be resolved pursuant to existing law, without the need for legislative change,” and the USPTO has reached a parallel conclusion on the patent side. Whether that view holds as the technology matures remains to be seen, but for now the conversation is one that IP attorneys, software developers, and business leaders cannot afford to ignore.

AI coding tools are powerful, and like all powerful tools, they demand a clear-eyed understanding of the responsibilities that come with them. The developers and companies that take IP questions seriously now, investing in legal guidance and compliance infrastructure before problems arise, will be far better positioned to build on a solid foundation as this technology continues to mature and the law continues to develop around it.

Latest patent guidance: https://www.uspto.gov/subscription-center/2025/revised-inventorship-guidance-ai-assisted-inventions

Latest copyright guidance: https://www.copyright.gov/ai/