Can AI Write a Patent? 2026 Legal Status and Patent Office Guidelines

The dream of pushing a button and watching a fully formed, legally bulletproof patent emerge from a machine is a powerful one for founders who are weary of high legal fees and long drafting cycles. However, the gap between "generating text" and "securing an enforceable legal right" is still wide, and the rules on who counts as an inventor are strict.

The "Inventorship" Trap: Why the Machine Can't Own the Idea

As a founder, you are likely used to AI helping you write code, marketing copy, or emails. In those fields, the stakes of authorship are relatively flexible. In patent law, they are rigid. The core keyword for the next decade of intellectual property is Inventorship.

The legal position in the major patent systems is now settled: an inventor must be a natural person. In the DABUS cases, where Stephen Thaler tried to name his AI system as the inventor, the US Court of Appeals for the Federal Circuit ruled against him in Thaler v. Vidal (2022) and the US Supreme Court declined to hear an appeal in 2023. The EPO Boards of Appeal and the UK Supreme Court reached the same conclusion. The only grant came from South Africa, whose patent system does not examine applications on substance.

In my two decades of practice, I have seen many founders try to "automate" their R&D. The danger is simple: if no human actually conceived what a claim covers, there is no proper inventor for that claim, and you end up with a piece of paper that looks like a patent but may not hold up when you need it.

Current Stance of the USPTO and CNIPA

The rules changed again recently, and many articles still describe the old US position.

1. The USPTO's 2025 Revised Guidance

In February 2024 the USPTO issued inventorship guidance for AI-assisted inventions that applied the Pannu factors to the human contribution. In November 2025 it rescinded that guidance in its entirety and replaced it. Under the revised guidance:

• There is no separate standard for AI-assisted inventions. The traditional test applies, and conception is the touchstone.
• AI systems are instruments used by human inventors, comparable to lab equipment, software or research databases.
• The Pannu factors only matter when deciding whether several natural persons are joint inventors.
• An application that lists an AI system as an inventor will be rejected, and a priority claim to a foreign application naming an AI tool as the sole inventor will not be accepted.

Simply asking an AI to "solve this problem" and copying the result leaves you unable to say what you conceived, which is exactly where the risk sits.

2. CNIPA: Inventors Must Be Natural Persons

The China National Intellectual Property Administration (CNIPA) published guidelines on AI-related patent applications at the end of 2024 stating that inventors must be natural persons and that AI systems cannot be named. For AI-assisted inventions, the natural person who made a creative contribution to the substantive features of the invention can be named. China's revised Patent Examination Guidelines, in force from January 1, 2026, add that inventors must be individuals (natural persons), that AI names may not be entered, and that false inventors may not be listed.

3. Honesty Obligations

In the US, everyone substantively involved in an application owes a duty of candor to the USPTO, and getting inventorship wrong can undermine the patent. China's Patent Law requires applications to follow the principle of good faith. Hiding how an invention was really made in order to get around inventorship rules is a risk in both systems.

Why "Raw" AI Output Faces High Rejection Risks

If you take a raw output from a standard Large Language Model (LLM) and file it as-is, you are likely to face several technical grounds for rejection that have nothing to do with the "AI-ness" of the document and everything to do with the quality of the drafting.

• The Enablement Gap: AI often "hallucinates" technical details. If your patent describes a process that doesn't actually work according to the laws of physics or engineering, it fails the "enablement" requirement. A patent must teach someone skilled in the art how to make and use the invention.
• Functional Claiming Traps: AI tends to describe what an invention does rather than what it is. This leads to "functional claiming," which examiners frequently reject for being overly broad or indefinite.
• Lack of Antecedent Basis: Patent drafting requires a very specific linguistic structure. If you mention a "widget" in the middle of a claim but didn't introduce "a widget" at the beginning, the claim is indefinite. AI models frequently lose track of these formal requirements over long documents.

The Strategist's View: A patent isn't a technical manual; it is a legal boundary. AI is excellent at writing manuals but struggles to understand where your legal boundary ends and your competitor's freedom begins.

How to Use AI for Efficiency Without Risking Your Assets

The goal for a modern business operator shouldn't be to avoid AI, but to use it as a "power tool" under the guidance of a master carpenter. You can cut drafting time by using AI patent drafting tools that are specifically built for legal compliance rather than general-purpose chat.

To keep your filings safe, follow these three principles:

1. Human-Led Claim Construction: Never let the AI decide what the claims are. You or your strategist should define the core inventive concept. Use the AI to help write the detailed description of what you have already worked out, or to describe alternative ways your team has identified for carrying it out.
2. The Conception Check: For every claim in your application, ask: "If I had to testify in court, what did I conceive here?" If the answer is "I typed a prompt," that claim is at risk.
3. Technical Verification: Every technical statement generated by an AI must be verified by your engineering team. If the AI suggests a specific chemical concentration or a specific data-processing architecture, you must ensure it is technically sound and actually reflects your R&D.

By treating AI as an assistant rather than the author, you gain the speed of modern technology while maintaining the legal integrity required to defend your business in court.