In June 2026, Anthropic disclosed something that should have stopped every engineering leader mid-scroll. As of May 2026, more than 80 percent of the code merged into Anthropic's production codebase was authored by Claude — up from low single digits when Claude Code launched in February 2025.
Anthropic's disclosure is a preview of where the industry is heading. Who legally owns AI-generated code remains less settled than most developers assume, and the answer depends on three things that have nothing to do with how good the code is: whether a human made enough creative decisions to establish copyright, whether an employment contract already assigned it to the employer, and whether the model pulled from GPL-licensed training data and contaminated the codebase without anyone noticing.
The American Baseline: Human Authorship or Nothing
The U.S. position is clear, if uncomfortable. Copyright protects work created by a human. The Copyright Office has confirmed this consistently, and the D.C. Circuit upheld it in Thaler v. Perlmutter.
On March 2, 2026, the Supreme Court declined to hear Thaler's appeal. Cert denial is not an endorsement. It means the Court chose not to hear the case — nothing more. What it does mean is that the D.C. Circuit's ruling stands, the Copyright Office's position is intact, and no court has gone the other way.
The practical implication is stark. AI-generated code accepted without meaningful human modification may not be copyrightable by anyone, leaving it effectively in the public domain. If a competitor copies it, there may be no legal recourse.
There is a narrow path to protection. Works that combine AI-generated material with sufficient human creativity can qualify — if the human selects and arranges AI outputs creatively, substantially modifies or edits the content, or integrates it into a larger human-authored work. But the Copyright Office has been explicit: "prompts alone do not provide sufficient human control to make users of an AI system the authors of the output". The Office has repeatedly rejected the argument that effort alone — including iterative prompt refinement — creates authorship. Prompts essentially function as instructions that convey unprotectable ideas.
The Open-Source License Risk
The copyright question is only half the problem. The other half is license contamination.
AI models are trained on massive public code corpora, including repositories under the GNU General Public License. GPL is a copyleft license: any derivative work must be distributed under the same terms. If an AI coding assistant generates a snippet derived from GPL-licensed code, and a developer merges it into a closed-source product, the result can be a "viral" license effect — a legal obligation to open-source the entire proprietary application.
The problem is growing. Open source licensing conflicts hit an all-time high in early 2026, partly driven by "license laundering" — AI assistants generating code snippets derived from copyleft sources without retaining the original license information. Black Duck's 2026 OSSRA report found that 68 percent of commercial codebases contained license conflicts, an all-time high in the report's history. Traditional software composition analysis tools scan declared dependencies. They miss AI-generated code that is structurally similar to GPL-licensed projects without directly importing them.
This is why the GNU Compiler Collection announced in July 2026 that it would reject any substantial code contribution generated by AI. The steering committee invoked the GNU maintainer guidelines' definition of "legally significant" contributions — roughly 15 lines of code or text — as the threshold for what requires copyright verification. If a contribution is more than trivial, and the training data cannot be verified, the contributor may not have the rights to share the patch under the GPL.
Linux took a different path. In April 2026, after months of debate, Linus Torvalds ended the argument with a characteristically pragmatic ruling: AI is a tool, use it if you want, but the human who clicks submit is responsible for any bugs, security issues, or legal problems that result. The policy does not ban AI. It requires disclosure and assigns accountability. AI-generated code must carry an "Assisted-by" tag and cannot use the standard "Signed-off-by" label, which carries a legal certification of origin.
GCC banned. Linux gated. Both are open-source projects confronting the same legal uncertainty, and they reached opposite conclusions.
The GitHub Copilot Case: What Survived
The most consequential AI code copyright lawsuit is also the least conclusive.
In November 2022, a group of programmers sued GitHub, Microsoft, and OpenAI, alleging that Copilot was trained on their open-source code without complying with license terms. The complaint alleged copyright infringement, DMCA violations, breach of open-source license terms, and other claims — 22 in total.
Judge Jon S. Tigar largely sided with the defense, dismissing 20 of the 22 claims. The biggest loss for the plaintiffs was the DMCA Section 1202 claim, which alleged that Copilot stripped copyright management information. The court found that the provision doesn't straightforwardly apply when AI generates code that lacks attribution — the AI isn't "removing" attribution in the traditional sense. It's generating new output that doesn't include it.
The copyright infringement claims also fell. The plaintiffs hadn't adequately alleged that specific copyrighted works were reproduced in specific Copilot outputs. AI code generation involves transformation and recombination, making it hard to draw a direct line from input to output without more specific allegations.
Two claims survived: open-source license violation and breach of contract. These rest on the argument that by using code governed by open-source licenses, GitHub and its partners agreed to the license terms, and Copilot's output violates those terms by failing to include required attribution and license text.
The case continues on those narrower grounds. The DMCA claim is on interlocutory appeal to the Ninth Circuit, and the outcome will set the attribution threshold for AI-generated code. But the central question — whether AI-generated code infringes copyright — remains unresolved.
China's Approach: Investment Over Authorship
China's legal framework for AI-generated content has developed in parallel, with a different emphasis.
In September 2026, the Supreme People's Court issued its first judicial guidance on AI-related disputes. The document explicitly states that AI-generated content alleged to infringe copyright cannot be exempted from liability simply because it was generated by AI. Responsibility must match the degree of control and duty of care.
But China has also developed a framework for protecting AI-assisted works. In a case decided by the Sichuan Free Trade Zone People's Court, a Shenzhen media company created six videos using generative AI. The court found that the company's process — writing scripts, generating storyboards, filtering materials, generating voiceovers, and editing — involved "continuous intellectual investment and substantive expressive transformation." The works were protected as audiovisual works, and the court ordered the defendant to pay compensation for infringement.
The standard is "continuous intellectual investment and substantive expressive transformation" — a test that focuses on the human's contribution to the creative process, not on who or what produced the final artifact. In China, as in the U.S., pure AI generation without human involvement is not protectable. But the threshold for human involvement may be lower in practice, and the emphasis is on the creative process rather than the authorship status.
The two frameworks produce similar outcomes in many cases. But they start from different premises. The U.S. asks: was there a human author? China asks: was there sufficient human investment?
What It Means for Builders
The practical guidance for engineering teams is not complicated, but it is uncomfortable.
Document human contributions. The more a team can demonstrate that a person made meaningful creative decisions — selecting, arranging, modifying, integrating — the stronger the copyright claim. A creative audit trail is not bureaucracy. It is asset protection.
Assume AI-generated code is uncopyrightable. If the code matters to the business, someone needs to modify it. Raw AI output is not an asset. It is a liability waiting to be copied.
Audit for GPL contamination. Traditional dependency scanners miss AI-generated code that is structurally similar to copyleft sources. The risk is not hypothetical. It is the reason GCC issued a blanket ban and the reason licensing conflicts hit record highs in 2026.
Disclose AI use. Linux requires it. The U.S. Copyright Office requires it for registration. The direction of travel is toward transparency, not concealment.
The legal frameworks in the U.S. and China are converging on a similar conclusion: AI-generated code is not automatically protected, and the burden of establishing ownership falls on the humans who deploy it. The difference is in emphasis — America asks who authored it, China asks what the human invested.
In both jurisdictions, the code that AI writes remains unprotected until a human establishes ownership through meaningful contribution.
Sources: American Bar Association "If AI 'Creates' It, Who Owns It?" (March 20, 2026); Shumaker, Loop & Kendrick LLP Client Alert (July 27, 2026); Neal & Leroy "What the Copyright Office's 2025 AI Reports Mean for 2026" (January 26, 2026); O'Reilly "Who Owns the Code Claude Wrote?" (June 15, 2026); The Next Web "Claude writes 80% of its code" (June 5, 2026); AI Vortex "Doe v. GitHub: What the Copilot Case Means for Legal AI" (April 11, 2026); Black Duck 2026 OSSRA Report via SD Times (February 26, 2026); GCC AI Policy Announcement (July 29, 2026); LWN.net "GCC steering committee bans AI-generated code" (July 29, 2026); ITHome "Linux AI Code Policy" (April 12, 2026); Legal Daily "Handling Intellectual Property Disputes in Line with the Law" (July 8, 2026); Supreme People's Court "Opinions on the Lawful Trial of AI-Related Dispute Cases" (September 7, 2026).
Disclaimer
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations
This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.
Forecasts from third-party analysts can change with market conditions.
Cost and pricing examples are point-in-time estimates; actual rates vary.
Country and company comparisons rely on public reporting, not operational data.
This sector moves fast; timelines and deal terms may be updated later.
Company deals and regulatory rulings may evolve; verify current status.
AI infrastructure is changing quickly; claims can become outdated soon.
Sources
- American Bar Association "If AI 'Creates' It, Who Owns It?" (March 20, 2026)
- Shumaker, Loop & Kendrick LLP Client Alert (July 27, 2026)
- Neal & Leroy "What the Copyright Office's 2025 AI Reports Mean for 2026" (January 26, 2026)
- O'Reilly "Who Owns the Code Claude Wrote?" (June 15, 2026)
- The Next Web "Claude writes 80% of its code" (June 5, 2026)
- AI Vortex "Doe v. GitHub: What the Copilot Case Means for Legal AI" (April 11, 2026)
- Black Duck 2026 OSSRA Report via SD Times (February 26, 2026)
- GCC AI Policy Announcement (July 29, 2026)
- LWN.net "GCC steering committee bans AI-generated code" (July 29, 2026)
- ITHome "Linux AI Code Policy" (April 12, 2026)
- Legal Daily "Handling Intellectual Property Disputes in Line with the Law" (July 8, 2026)
- Supreme People's Court "Opinions on the Lawful Trial of AI-Related Dispute Cases" (September 7, 2026).
The information provided in this article is for general informational and educational purposes only. It does not constitute legal, financial, or professional advice. The author and publisher are not responsible for any actions taken based on the content of this article. Readers should consult qualified professionals for advice specific to their situation. All trademarks and references to third-party products, services, or organizations are the property of their respective owners. The performance data and benchmarks discussed are based on specific research studies and may not generalize to all use cases or environments. As of the publication date, the AI landscape continues to evolve rapidly, and readers should verify current information independently.
Limitations: This analysis is based on reporting and public data available as of the article date; figures may be revised as sources update.; Forecasts from third-party analysts can change with market conditions.; Cost and pricing examples are point-in-time estimates; actual rates vary.; Country and company comparisons rely on public reporting, not operational data.; This sector moves fast; timelines and deal terms may be updated later.; Company deals and regulatory rulings may evolve; verify current status.; AI infrastructure is changing quickly; claims can become outdated soon.