Patenting software after Alice, and what the AI inventorship cases mean
Software patents are still available, but the framework has narrowed and the AI inventorship question has now been answered.
The question I get most often from technical founders is whether their software is patentable. The short answer is that it can be, but the framework that controls the answer is narrower than most founders assume, and the answer for code generated with substantial assistance from an AI system is narrower still.
The controlling case is still Alice Corp. v. CLS Bank, decided by the Supreme Court in 2014 and still doing most of the work in software patent litigation a decade later. Alice established a two-step test. First, the court asks whether the claim is directed to a patent-ineligible concept (an abstract idea, a law of nature, or a natural phenomenon). If yes, the court then asks whether the claim contains an inventive concept sufficient to transform the abstract idea into a patent-eligible application. Most software claims fail at step two, because reciting “do this abstract idea on a computer” or “do this abstract idea over the internet” is not enough to clear the threshold. The Federal Circuit has applied Alice to invalidate hundreds of software patents in the years since.
What survives Alice is software claims that solve a technical problem in a non-obvious way and that are described in claims tied to specific implementations. A pure business-method patent that happens to be implemented in code is the weakest position. A claim describing a specific improvement to how a computer functions (faster, more secure, less resource-intensive, accomplishing something that was not previously possible) is the strongest. Drafting the application is where the case is won or lost. The same underlying invention can be patentable or unpatentable depending on how the claims are written and how the specification frames the technical contribution.
The second variable, which is newer, is AI inventorship. In Thaler v. Vidal, the Federal Circuit held in 2022 (and the Supreme Court declined to review in 2023) that an inventor under the Patent Act must be a natural person, not an AI system. The USPTO followed with guidance in 2024 confirming that AI cannot be a named inventor, but that AI-assisted inventions can still be patented as long as a natural person made a significant contribution to the conception of the claimed invention. That guidance has practical teeth. If a software invention is largely the output of an AI coding assistant, with limited human contribution beyond prompting, the patent may be unenforceable for failure to name the actual inventor (because the actual inventor is not a person and cannot be named at all). Engineering organizations relying heavily on AI-assisted development should document where the human conceptual contribution lies, and should treat that documentation as part of the patent record from the outset.
The “first to file” rule from the America Invents Act is the third variable that founders consistently underestimate. Two engineers working independently on the same idea will produce different patent outcomes based on filing date, not invention date. The startup that delays filing because the product is not finished can lose the patent to a competitor who filed a thinner application earlier. Provisional applications are the standard tool here, and they are inexpensive relative to the risk they hedge.
The practical takeaway for software founders is to make three decisions early. First, decide which technical innovations in the codebase are candidates for patent protection and which are better protected as trade secrets, because the choice is mutually exclusive once a patent application publishes. Second, if patent protection is the right tool, get to a provisional filing quickly and let the priority date do its work. Third, if AI-assisted development is part of the engineering process, build the inventorship record contemporaneously so the application can survive a later challenge. The patent system was designed for a slower, more linear style of invention than software development now operates on. The work-arounds exist, but they only work if you set them up before you need them.