A federal appeals court just settled a question every M&A lawyer should have been asking for two years: can a machine own what it creates? The answer is no. And that answer has quiet, expensive implications for anyone buying a company built on AI-generated code. Let’s enter Thaler v. Perlmutter.
The Case
In March 2025, the D.C. Circuit ruled on Thaler v. Perlmutter. Dr. Stephen Thaler tried to register a copyright for an image his AI system, the “Creativity Machine,” produced on its own. He listed the machine as the sole author and himself only as owner. The Copyright Office refused registration. The district court agreed. The Court of Appeals affirmed all of it.
Why Authorship Requires a Human
The court’s reasoning goes deeper than a simple policy preference for humans over machines. It’s a textual argument, built provision by provision through the Copyright Act of 1976.
Copyright vests in an author immediately upon creation. It lasts for the author’s life plus seventy years. It passes to a spouse or children on death. It can only be transferred through a signed instrument. Every one of these mechanisms presumes a legal person with a lifespan, a family, and the capacity to sign a document. A machine has none of that. Read literally, the statute breaks down the moment you substitute “machine” for “author.”
The court also leaned on history. The Copyright Office adopted a human-authorship rule back in 1973, three years before the current Act passed. Congress studied the issue directly through the CONTU commission in the mid-1970s and concluded that computers are tools, not creators. When Congress reenacted the word “author” in 1976, the court treated that as an implicit adoption of the interpretation already settled by the agency and by CONTU. That’s a meaningful legal move: it means the human-authorship rule isn’t a modern judicial invention reacting to generative AI. It’s read as baked into the statute from the start.
What the Ruling Does Not Say
This is the part that gets lost in headlines. The court explicitly did not hold that AI-assisted work is uncopyrightable. A human who directs, edits, or meaningfully shapes AI output can still be an author. The Copyright Office’s own 2023 guidance says as much: registrability depends on how the tool was used, not on whether a tool was used at all.
The line the court drew sits between assistance and autonomy. Thaler lost because he affirmatively told the Copyright Office, repeatedly, that the work was created with no human creative input whatsoever. That was a fact he chose to put on the record. A different applicant, with a different record, could reach a different result.
Where This Lands on Code
Software sits closer to that line than most people realize. AI coding assistants now generate entire functions and modules with minimal human review. Nobody is documenting how much human judgment went into any given block of code, because until this ruling, nobody had a legal reason to.
That gap is a title problem. If a company can’t establish that a human meaningfully authored the code in its repository, its claim to own that code weakens, regardless of what any employment agreement or IP assignment says. An assignment can only transfer rights that exist. If the underlying work was never eligible for copyright in the first place, there’s nothing to assign.
This compounds an existing due diligence blind spot: open source code entering proprietary repositories through AI-assisted development, often undetected by standard dependency scanners. Both issues share the same root problem. Code is being produced faster than anyone is tracking its origin or its legal status.
Who should bear the burden of proving human authorship in a deal: the seller making the representation, or the buyer relying on it?
This article was written with the help of AI.
Are you preparing for an acquisition or investment round? At Fossity, we help companies audit their open-source software early, reduce risk, and build trust with buyers. Contact us today to strengthen your M&A readiness.