CommentaryA junior associate at a mid-sized law firm asks an AI tool for a first draft of a shareholder agreement. She spends an hour tightening the definitions and rewriting two clauses, then sends the result to a partner. The client pays the invoice. Nobody asks who owns the draft, because the answer seems obvious. The firm paid her salary, so the firm owns her work. That answer is probably right for the hour she spent editing. For everything the tool wrote before she touched it, nobody can say with confidence. In several major jurisdictions, the best reading of the law is that nobody owns it at all.
The Rule Assumes an Author
Employer ownership rules were built on a simple chain. A human creates a work, and the law then decides whether the creator or the employer holds the rights. Under US law, a work prepared by an employee within the scope of employment is a “work made for hire,” and the employer is treated as its author. In the United Kingdom, section 11(2) of the Copyright, Designs and Patents Act 1988 makes the employer first owner of a work made by an employee in the course of employment, unless the parties agree otherwise. Both rules decide who gets the rights. Neither says whether any rights exist. That is a separate question, answered by the test of authorship, and generative AI puts that test under strain. If the employee did not author the relevant text, the employer-ownership rule has nothing to pass along. An employer cannot inherit what the employee never held.
Four Systems, Four Answers
The United States has been the clearest. In Thaler v. Perlmutter, the US Court of Appeals for the District of Columbia Circuit held in March 2025 that the Copyright Act requires a human author. The US Supreme Court declined to review the case on March 2, 2026. A refusal to hear a case is not a ruling on the merits, and Thaler was the easy case. Stephen Thaler named his own AI system as the sole author and disclaimed any human creative input. The court did not decide how much human involvement is enough when a person works with AI. The Copyright Office has given a partial answer. Its January 2025 report on copyrightability concluded that, with current technology, prompts alone do not give a user enough control over the output to make the user its author. Human selection, arrangement, and modification can still be protected, and so can human-written material that remains visible in the output. For our associate, that means her edits are probably protected and the machine’s first draft probably is not. Her firm holds rights in the editing and a doubtful claim to the rest.
The United Kingdom took a different route in 1988. Section 9(3) gives copyright in a computer-generated work with no human author to the person who made the arrangements necessary for its creation, for 50 years. For decades it looked like a clever fix. In March 2026 the UK government’s report on copyright and AI said the provision should probably be removed, after a clear majority of consultation respondents said works made solely by AI should not be protected. Protection for works made with AI assistance would stay. The report set no timetable. A firm relying on section 9(3) today is relying on a provision its own government wants to remove.
Elsewhere the picture is mixed. EU law protects works that are the author’s “own intellectual creation,” the standard from the Court of Justice of the European Union’s Infopaq decision, which points toward a human author. In China, the Beijing Internet Court held in November 2023 in Li v. Liu that an image made with Stable Diffusion was protected. The court found that the user’s choices and adjustments reflected his own intellectual input, and it declined to give the rights to the software’s developer. So the same machine-made paragraph may be protected in Beijing, doubtful in Washington, and protected in London for now but probably not for long. A firm with offices in all three cannot give a client a single answer about who owns a deliverable.
The Leak Nobody Drafted For
Ownership is the quieter problem. Confidentiality has already caused damage. In early 2023, engineers in Samsung’s semiconductor division reportedly pasted internal source code into ChatGPT to fix bugs, and one reportedly used it to summarise meeting notes. Within weeks Samsung banned generative AI tools on company devices and networks, with disciplinary action up to dismissal for breaches. The courts have now produced their own version. In United States v. Heppner, Judge Jed Rakoff of the US District Court for the Southern District of New York ruled in February 2026 that written exchanges a defendant had with a consumer AI platform, which he later shared with his lawyers, were covered by neither attorney-client privilege nor work-product protection, the US rule that shields materials prepared in anticipation of litigation. Among his reasons were that the platform is not a lawyer and that its privacy terms allowed disclosure of user inputs, so the exchanges were not confidential. The defendant had acted on his own initiative, and the judge expressly left open whether the result would differ if counsel had directed the use of the tool. This is one trial-court ruling, and it binds no other court. The point for employers is narrower and firmer. An employee who types client material into a public AI tool may be handing a third party something that was protected, and a court may treat the protection as gone. Most confidentiality clauses were written to stop employees talking to outsiders. They say nothing about typing into a text box. The employee who does it has not knowingly breached anything, and the contract never told her otherwise.
When the Draft Is Wrong
In Mata v. Avianca (2023), a federal judge in New York ordered two lawyers and their firm to pay $5,000 after they filed a brief citing cases that an AI tool had invented and then stood by those citations when the court questioned them. The decision is usually told as a story about careless lawyers. It is also a story about employment, because the firm was sanctioned alongside the individuals. In most common-law systems an employer is answerable for work done in the course of employment. The employer who never approved the tool will say the associate acted outside her instructions. She will say she used what everyone around her was using. Both are describing the same gap from opposite sides, and without a written policy a court has little to decide between them.
What the Contract Should Say
Courts and legislatures will take years. A contract can change next week. Four clauses would close most of the gap. First, approved tools. The contract or a policy it incorporates should name which AI systems employees may use and on what kind of account. A paid enterprise account that does not train on user data is a different risk from a free personal one. Second, confidential material. The clause should bar entering client or confidential information into unapproved systems and should define “entering” to include uploading documents, not only typing. Third, ownership and disclosure. Employees should assign to the employer every right they may hold in AI-assisted work, including rights in their selections, edits, and arrangements. They should also record which parts of a deliverable were machine-generated. An assignment cannot create rights that do not exist, but it captures every right that does, and it will matter if the law moves in the employee’s favour. Disclosure matters for a practical reason too. The US Copyright Office requires applicants to disclose AI-generated content and to disclaim any that is more than de minimis, and an employer who cannot trace the machine’s contribution cannot make that disclosure truthfully. Fourth, responsibility. A named human should check facts, citations, and figures before any AI-assisted work leaves the building, and the contract should say who that is. None of this needs a statute or a test case. It needs someone to sit down and write it.
The Promise a Firm Cannot Yet Make
Clients continue to ask their advisers one question: who owns what you deliver? Today an honest answer has three parts. The human contribution is protected. The machine’s contribution may not be, depending on the country. And the firm can only say which is which if it recorded the difference when the work was done. The law will eventually settle who owns what a machine helped write. Until it does, the people drafting employment contracts are settling it by default, whether they realise it or not.
Sheraz Ahsan is an LL.B. (Hons.) candidate at the University of London. He writes opinion pieces for The Friday Times and The News International and has worked as a legal intern at Cornelius, Lane & Mufti in Lahore. His research interests include employer-employee intellectual property ownership, defamation law, and corporate and securities matters.