AI-Driven Design and IP Laws

Grant Steyer, a patent attorney and a partner at the IP Law Firm Renner Otto, usually works with companies as well as individual inventors. He has been closely watching the emergence of IP-driven design tools. In this podcast episode, he addresses the legal implications of working with generative design tools.

Current AI tool usage fundamentally reflects how engineers have been working all along, Steyer points out. “It’s just like what you would have done in the past—Googling, going on Reddit, talking to a colleague.” The problem he often sees is patent application drafts with chunks of AI-generated gobbledygook.

“Reading through the writeup, I’d see parts that I can’t figure out. So I’d go to the inventor to talk about it, and he’d say, ‘Oh yeah, that one I wasn’t quite sure, but ChatGPT was convincing, so I left it in there.’ And it’s something that basically breaks physics—it’s not possible,” he says. His suggestion is to use AI “like an assistant or intern who is giving you some good work, but doesn’t have a lot of experience, so you know you shouldn’t trust everything he or she gives you.”

Can AI be an Inventor?

On November 28, while most of us were still recovering from our indulgent Thanksgiving meal, the U.S. Patent Office issued new guidelines for AI-assisted inventions. It reads, “AI systems, including generative AI and other computational models, are instruments used by human inventors. They are analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process ... “ It emphasizes that a human or humans must “contribute in some significant manner to the conception.” Citing the Federal Circuit, it states, “AI cannot be named as an inventor on a patent application (or issued patent) and that only natural persons can be inventors.”

Accidental Patent Infringement 

Because of how machine learning works, there is, however, a risk that “you might accidentally infringe on someone else’s patent without meaning to,” Steyer notes. This has to do with the data set employed by the software developer to train the AI system; specifically, whether that data set includes patented and copyrighted designs.

At Autodesk University, the company’s CEO Andrew Anagnost promoted Neural AI, a design tool that understands natural language input. As a safeguard against inadvertent design plagiarism, he says, “if something is generated by our model that looks like something in the training set, we throw that result away. That protects IP ...” (Read our Autodesk University 2025 here.)

Steyer sees Autodesk’s safeguard as “a good first step.” He adds, “I’d feel much better if I see a training data set that’s open source, or the developer has paid licensing fees for.”

For more on legal implications and possible pitfalls with AI design, listen to the full podcast. 

From your vantage point, how do you see inventors or engineers using AI in their workflow? And do you see any problem with the current way in which they are using it?

I’ve seen some good and some bad. A lot of the inventors now are using generative AI to kind of come up with different embodiments, I would call them, for their invention, or to come up with ways to address particular problems with the invention. For example, if there’s a client that is using pneumatics, compressed air to deliver a sample, and there’s some issue with changing the orientation, they can use generative AI to propose some different solutions. 

It’s the sort of thing that you would have done in the past by Googling, going on Reddits, talking to a colleague and trying to get the information, and generative AI gives you a bunch of different answers that you can go through quickly. The problem I’ve seen is that sometimes, and I think this goes for everybody, it is assuming or believing the output a little too much. when you assume anything’s accurate that’s coming out of these generative AIs is when you can get into some trouble. 

A number of CAD software and simulation software companies we regularly write about have introduced AI features into their design software, so you can use a natural language prompt to generate a shape or design. This raises a question about patent applications, where there needs to be a significant contribution by a human. Does this AI trend put that “human contribution” in jeopardy?

The U.S. Patent and Trademark Office (USPTO) put out some guidelines for determining inventorship when AI is involved. They put out new ones on November 28 [2025]. The human has to conceive of the invention; if all of the key features were [generated] by the AI, then that’s a problem. Normally the patent office is just treating it as a tool. More often than not, you have a human in the loop working with the AI.

The AI is only really a problem when there is really no human involved. That’s not something I’m seeing or coming across right now, but that’s one of the fears. If it eventually gets to that point, then you may have a problem with inventorship.

The way current design software works with AI tools is that they have to rely on existing training data and designs. One of the safeguards that some of the leading software developers have implemented is that if the AI spits out something that looks identical to something in the training data set, they discard that design. Is that a sufficient safeguard? What are your thoughts on that approach?

There are basically two main issues with generative AI in that sort of situation. One is that you accidentally infringe someone else’s patent without meaning to. On the other side of that, if you have all of these what I could call “innocent infringers,” as a business owner it’s going to be annoying to kind of play whack-a-mole and send cease and desist letters to go after these people infringing your patents.

I see [those safeguards] as a good first step. But there may need to be a way where the designers of these CAD files are compensated. Maybe they’re paid a license fee to be [in the training data.] Maybe that’s the way it ultimately works out, because it seems it can be very difficult to have these software checks that are good enough at looking for potential infringement.

What are your tips for engineers and designers that want to test out these generative AI tools, and get to the point where they need to submit a patent?

One thing, and I kind of mentioned this earlier, is having a human in the loop, and using the generative AI almost like a coworker, a tool, something that you’re going back and forth with. You’re not asking the generative AI to output a completed design based on a problem statement without a solution involved. The main issue I’ve seen is when people trust too much what is output by the generative AI, because there are normally problems with it that they might not be aware of. With the kind of hallucinations we see with these generative AIs, for example, when asking it to cite to different parts of a patent, it makes up citations all the time. And if it’s having trouble with that, it’s hard for me to believe there’s not going to be other issues there. 

You can also use generative AI to supplement your skill sets for doing things more quickly; even for things like doing research or pulling citations. You can also use it for record keeping. Most of these generative AI tools keep a log of these prompts. If you output that and store it somewhere, it can help prove you came up with the invention. 

Also remember to turn all the privacy to keep your data safe. Make sure those options are turned on within the tools to keep your data and your work private as much as possible.