Engineers now routinely use generative AI tools to sketch designs, propose alternatives, generate code, simulate outcomes, and identify gaps in their own thinking. This is changing how invention happens — and it is raising uncomfortable questions that the patent system has not fully answered. Who is the inventor when a large language model contributed substantively to the solution? Does AI involvement affect whether a patent can issue at all? What should engineering teams do differently to preserve their rights?
The short answer is that AI-assisted inventions are patentable, but the AI cannot be named as an inventor, and the humans who are named must have contributed more than just running a prompt. The longer answer, and the practical guidance for teams actually shipping work, is below.
Only Humans Can Be Inventors
The Federal Circuit settled the threshold question in Thaler v. Vidal (2022): under the Patent Act, an inventor must be a natural person. An AI system cannot be listed as an inventor on a U.S. patent, full stop. The Supreme Court declined review. The USPTO's 2024 inventorship guidance confirmed the same position and applied it agency-wide.
This does not mean AI-assisted inventions are unpatentable. It means that for every claim in your patent, at least one natural person must qualify as an inventor under the traditional legal standard — which means they must have made a "significant contribution" to the conception of the claimed subject matter.
The 2024 USPTO Guidance: What "Significant Contribution" Means
In February 2024, the USPTO issued guidance specifically addressing AI-assisted inventions. The guidance applies the traditional Pannu factors to AI-assisted work: a person qualifies as an inventor if they made a significant contribution to the conception of the invention, did not contribute only in a manner well-known or insignificant, and did more than simply explain the state of the art. The guidance then lays out principles for evaluating whether a human's role around an AI tool satisfies this standard.
The principles most relevant in practice:
- Merely recognizing a problem or a general goal is not enough. Telling the AI "design a better heat exchanger" does not make you an inventor of whatever the AI returns.
- Reducing an AI-generated output to practice is not, by itself, an inventive contribution. Building and testing what the AI gave you does not make you the inventor of the concept.
- Eliciting a specific output through targeted prompting and iteration can contribute significantly, especially when the human selects, modifies, and refines results in ways that require technical understanding.
- Designing the AI system itself, its training data, or the specific constraints under which it operates can be an inventive contribution to the output produced, depending on the causal relationship between those design choices and the claimed invention.
"The right question isn't 'did AI help?' It's 'did a human make a significant contribution to the conception of what's claimed?' If yes, name that human. If no, you don't have a patentable invention — regardless of how impressive the AI output was."
Ownership Is a Separate Question
Inventorship and ownership are often confused but are legally distinct. Inventorship is a question of who contributed to conception. Ownership is a question of whose rights flow from that contribution — typically governed by employment agreements, assignment clauses, and contracts.
Ownership problems specific to AI-assisted work:
- AI vendor terms of service. Some AI platforms claim broad licenses to user inputs and outputs. If your engineers are pasting proprietary technical descriptions into a public AI tool, you may be inadvertently granting your vendor rights you do not want them to have — and potentially creating prior art that destroys novelty. Read the terms for every tool your team uses.
- Contractor and consultant work. If an outside contractor used AI to generate work product for you, your assignment from them transfers only what they actually own. If their AI tool's terms assigned rights elsewhere, you inherit that gap.
- Training-data contamination. If an AI tool was trained on your competitor's publicly disclosed technology and produces output that reads on that technology, you may be obtaining a patent application that infringes something else. A freedom-to-operate analysis is especially worth running for heavily AI-assisted inventions.
Documentation Practices for AI-Assisted Work
The single most valuable habit an engineering team can adopt is documenting the human contribution. In a USPTO dispute or a future litigation, the question will be what each named inventor actually did. Contemporaneous records win those arguments.
Practical documentation to keep:
- The problem statement, as articulated by the human before any AI tool was involved.
- The prompts and constraints used to direct the AI, with attribution to the individual who authored them.
- A record of what the AI produced and what the human did next — rejected, modified, combined with other outputs, tested against specific criteria, generalized into a broader approach.
- The decisions the human made that the AI did not make: which alternatives to pursue, which parameters to vary, which failures to treat as signal.
This is not a bureaucratic exercise. This is the evidence supporting the inventorship named in the application. It also protects against a later discovery that a contributor was omitted — a defect that can invalidate a patent if done with deceptive intent.
A Word on Disclosure to the USPTO
The USPTO guidance notes that applicants have a duty of candor, which includes disclosing information material to inventorship. If AI contributed significantly to the claimed invention, the safer path is to be prepared to answer questions about the human contribution. You are not required to volunteer a confession, but you cannot hide facts the examiner is entitled to consider.
Key Takeaways
- AI cannot be named as an inventor. At least one human must qualify under the traditional significant-contribution standard.
- Recognizing a problem or reducing an output to practice is not inventorship. Meaningful technical direction, selection, modification, or system design often is.
- Check the terms of service for every AI tool your team uses. Vendor terms can silently undermine ownership and create prior-art risks.
- Document the human contribution contemporaneously. Prompts, decisions, and modifications are the evidence that proves inventorship later.
- Treat AI-assisted inventions as a reason to tighten your freedom-to-operate practice, not relax it.
Next Steps
If your team is using AI in the invention process and you want a practical protocol that preserves patent rights without slowing engineers down, schedule a consultation. Most organizations already have the facts they need — what's missing is the documentation habit that converts those facts into defensible patents.
