AI and IP Protection Across Borders: How the US, EU, China and UK Treat AI-Generated Work

Artificial intelligence has forced every major intellectual property system to confront a question their statutes were never written to answer: can output produced by, or with heavy help from, a machine be owned, and if so, by whom? There is no treaty that settles this, and no international consensus. Instead there is a patchwork of statute, office guidance and court decisions that pull in different directions and are changing quickly. This overview sets out, at a high level and without giving advice, how four of the most-cited systems (the United States, the European Union, China and the United Kingdom) currently approach AI across the two areas where the tension is sharpest: copyright authorship and patent inventorship. For the wider framework these questions sit inside, see the international IP hub.

Why there is no single answer

Two things make AI and IP hard to generalise. First, the label "AI-generated" hides a spectrum. A work can be almost entirely machine-produced from a short prompt, or it can be a human creation in which AI played a minor, tool-like role, with countless gradations between. Systems tend to treat those ends of the spectrum very differently. Second, most IP laws were drafted around an assumption of human creativity and human invention, so applying them to autonomous systems means either reading old words in new ways or waiting for legislators to act. Different jurisdictions are doing this at different speeds, which is why positions that look settled in one country remain open in another.

Patents: most offices require a human inventor, but the boundaries are unsettled

On the narrow question of whether an AI system can itself be a named inventor, there is more convergence than divergence, though it is not unanimous. The high-profile test cases brought under the DABUS project, which tried to name an AI system as the inventor, have generally failed. Patent offices and courts in the United States, at the European Patent Office, and in the United Kingdom have taken the position that a named inventor must be a natural person, not a machine. A number of other major offices have reached similar conclusions, but the picture is contested rather than uniform: a small number of jurisdictions have diverged or initially diverged. South Africa, which runs a formalities-only system without substantive examination, granted a DABUS patent, and in Australia the Federal Court at first instance held that an AI could be named as inventor before the Full Federal Court reversed that finding. China's patent office, CNIPA, is not analysed in detail here, so its position on AI inventorship should be checked separately rather than assumed from the others.

That the machine itself cannot be named does not mean inventions made with AI assistance are unpatentable, and this second question is genuinely open. The prevailing approach is to ask whether one or more human beings made a sufficient contribution to the claimed invention. Where a person has meaningfully directed, shaped or refined what the AI produced, that person can usually be named, and the invention assessed on the normal patentability criteria (novelty, inventive step and industrial application or, in the US, utility). Where no human contribution can be identified, protection becomes doubtful. Office guidance on exactly how much human input is enough is still developing and differs in emphasis between jurisdictions, so this is an area to confirm on the current position rather than assume. You can read more about the systems referenced here on the United States, European Union and China jurisdiction pages.

Copyright: this is where systems diverge

Copyright is where the international picture fragments. The core disagreement is whether a work with no human author, or only a thin layer of human involvement, can attract protection at all.

The United States takes a firmly human-authorship line. The US Copyright Office has repeatedly stated in its guidance that copyright protects the products of human creativity, and that material generated purely by AI, without creative human control over the expressive result, is not registrable. Works that combine human authorship with AI-generated elements can be registered, but generally only for the parts a human actually authored, with AI-generated portions disclaimed. Separately, litigation has reinforced that a machine cannot be the sole author. These are two distinct sources, office guidance and court decisions, rather than a single ruling on how human-plus-AI works are treated generally.

The European Union has no single harmonised rule specific to AI, but its case law sets a demanding threshold. To qualify for protection, a work must be the author's own intellectual creation reflecting free and creative choices. That standard is widely read as requiring a human author, which makes fully autonomous output difficult to protect, though member states apply the principle through their own national laws and the details vary.

China has moved in a more permissive direction in specific decisions, but not uniformly. Lower courts, including the Beijing Internet Court, have in individual first-instance cases recognised protection for AI-assisted output where a human user exercised enough intellectual input and creative choice for the result to count as original. Other Chinese decisions have declined protection on their own facts. These are court decisions on particular fact patterns rather than appellate or statutory doctrine, so they should be read narrowly and treated as an evolving and not-yet-uniform line, not a blanket national rule that China protects AI output.

The United Kingdom is the notable outlier among the four. UK copyright law contains a long-standing provision for computer-generated works that have no human author, deeming the author to be the person who made the arrangements necessary for the work to be created. This means the UK, on paper, offers a route to protect certain computer-generated output that other systems would refuse. How this older provision maps onto modern generative AI, and whether it survives in its current form, has been the subject of government consultation and debate, and its future is genuinely uncertain.

AI-assisted versus AI-generated: the line that matters most

Across all four systems, the single most important distinction is between AI as a tool and AI as the author or inventor. Work in which a human makes the meaningful creative or inventive choices, using AI to execute or accelerate, is far more likely to be protectable everywhere. Work that a system produces with only a trivial human prompt sits in the contested zone, protected in some places, unprotected in others, and often unprotected as to the machine-made parts even where the overall work can be registered. Documenting the human contribution to a project is therefore prudent regardless of jurisdiction.

Practical implications while the law settles

Because the rules differ and are in flux, an asset that is protectable in one country may be unprotectable in another, which matters for anyone operating across borders. Two knock-on issues are worth flagging. Ownership and chain of title can be murky when AI is involved, so contracts, employment terms and commissioning agreements should be clear about who owns what. Licensing is a related pressure point, both for the training data used to build AI systems and for the output they generate; the framing questions there are covered in the IP licensing overview. Timeframes for filing, registration and any consultation-driven law changes vary by system and shift over time, so treat any specific period as provisional and confirm the current window with the relevant national IP office or WIPO or local counsel.

A moving target

This is one of the fastest-moving corners of intellectual property law. Court decisions, office guidance and legislative consultations are appearing continually, and a position described as current today may be superseded. The safe assumption is that AI and IP is unsettled almost everywhere, that most offices lean towards requiring a human inventor while a few have diverged, that copyright remains genuinely split, and that the details must be checked afresh for each jurisdiction and each project.

IPEnvoy is not a law firm and does not provide legal advice. This is general information only. Confirm the current position with the relevant national IP office or WIPO and a qualified local IP professional before making decisions about protecting or commercialising AI-generated or AI-assisted work.