Who Owns the Code Claude Code Wrote? Legal Experts Still Can't Agree

A legal analysis published on Substack's Legal Layer tackled one of the most practically urgent IP questions in software right now: **if Claude Code writes your code, who owns it?** The post surfaced to the top of Hacker News with 499 points and hundreds of comments, suggesting the question has moved from theoretical to urgent for working developers.

The core tension: US copyright law requires human authorship for protection. The Copyright Office has consistently declined to register purely AI-generated works. But "Claude Code wrote this" rarely means zero human involvement — prompts, edits, reviews, and architectural decisions all inject human authorship back into the chain.

The analysis walks through several scenarios. **Pure generation** (paste prompt, accept output verbatim) is likely unprotectable. **Substantial human editing** creates a derivative work where the human-authored portions are protected. **Work-for-hire doctrine** applies if the code is created within an employment relationship — meaning the employer likely owns it regardless of how much AI contributed, which is a different kind of problem for open-source contributors using Claude Code on the side.

For companies, the risk is asymmetric: if your product is built substantially on unprotectable AI output, a competitor can legally copy it. For individual developers contributing to open source, the question of whether AI-assisted PRs comply with project licenses is unresolved.

Anthropicʼs own terms grant users ownership of outputs — but that doesn't solve the underlying copyright status under federal law, which no terms of service can override. The community consensus in the HN thread: **talk to a lawyer, document your human contributions, and watch the Thaler v. Vidal appeals court ruling**, which may clarify the human-authorship requirement in the coming months.