US Patent Office says AI can help, but only humans can be inventors

If your AI model independently designs a brand-new mechanical device from scratch, can you list that algorithm as a co-inventor on the paperwork? The U.S. Patent and Trademark Office just gave its official answer: absolutely not.

The USPTO published new guidance clarifying that only "natural humans" can hold patents. The office emphasized that patents exist fundamentally to incentivize human ingenuity, meaning AI systems cannot be credited as inventors or even joint inventors under current federal law.

The update comes directly in response to the Biden administration's executive order on AI from last fall, which mandated inventorship rules by late February. It also codifies the precedent set by last year's Thaler v. Vidal case, where a federal appeals court rejected an application that listed an AI system as the sole inventor.

The prompt engineer's patent bar

AI-assisted inventions are not dead on arrival. The USPTO made it clear that using generative models will not automatically disqualify your application, provided a human made a "significant contribution" to the core conception.

That creates a very tricky gray area. According to the guidance, simply tossing a general problem to an AI or setting up an abstract research goal does not count as conception. If an inventor asks a chatbot to design a critical part for a remote-control car and the bot does all the heavy lifting, that human cannot claim the patent.

Simply owning or overseeing the model also gives you zero legal credit over its autonomous outputs. However, the agency noted that crafting highly specific prompts to elicit a tailored technical solution, or taking raw AI output and making significant modifications to it, can legally qualify you as an inventor. Dorsey & Whitney attorney Jamie Nafziger pointed out that patent examiners now face a wild new burden: deciding exactly how sophisticated a prompt needs to be before it crosses the inventorship threshold.

Why it matters

We think this guidance gives corporate R&D teams much-needed breathing room. Enterprises sinking billions into AI for drug discovery, material science, and chip architecture now have a clear green light: keep using AI, but make sure your human engineers document their prompts, workflows, and subsequent modifications. For now, the patent pipeline remains intact.

Yet the agency left behind a glaring blind spot. The guidelines do not strictly require applicants to disclose whether AI was used during the invention process. As consumer advocacy group Public Knowledge pointed out, this creates an open invitation for patent trolls to quietly churn out floods of speculative, low-quality patent filings using automated models, only to weaponize them against productive businesses later.

The USPTO is drawing the exact same line as the U.S. Copyright Office by insisting on human agency above all else. It is a sensible stopgap for today, but as autonomous agents move from answering prompts to running end-to-end scientific experiments, defining where AI stops and human contribution begins is going to turn into a legal nightmare.