Why Software and AI Patent Claims Get Invalidated: The §112(f) Functional Claiming Trap

Short answer: Under 35 U.S.C. §112(f), a claim element that recites a function without sufficient structure can be treated as “means-plus-function” — even when the word “means” never appears. If the specification then fails to disclose corresponding structure (for software, typically an algorithm), the claim can be held indefinite and invalid. Software and AI claims using generic terms like module, unit, component, or handler are especially exposed. For companies building patent portfolios around software, algorithms, or AI systems, this is one of the most consequential drafting risks in US patent law — and it is almost entirely preventable at the drafting stage. What §112(f) actually does Section 112(f) lets a patentee claim an element as a means for performing a function, rather than by reciting its structure. The trade-off is scope: when it applies, the claim covers only the corresponding structure disclosed in the specification and its equivalents — not every way of achieving the function. The classic trigger is the phrase “means for.” But that is no longer the boundary. The “means” word is not the test Where a claim does not use “means,” there is a rebuttable presumption that §112(f) does not apply. That presumption can be overcome if the term recites function without reciting sufficiently definite structure to a person of ordinary skill in the art. In Fintiv, Inc. v. PayPal Holdings, Inc. (Fed. Cir. Apr. 30, 2025), the claims recited a “payment handler” and “payment handler service” configured to perform certain functions. The claims never used “means.” The Federal Circuit nonetheless held that §112(f) applied: “handler” conveyed no sufficient structure to a skilled artisan, and the term recited what the element did rather than what it was. Why the specification then becomes fatal Once §112(f) applies, the court looks to the specification for the corresponding structure. For a software-implemented function, that generally means an algorithm — the steps by which the function is actually performed. In Fintiv, the specification largely restated the claim language rather than explaining how the functions were implemented. The patentee argued a two-step algorithm was disclosed (wrapping the APIs of different payment processors, then exposing a common API), but the court found this merely repeated the claim language. Without an algorithm, and given the failure to disclose adequate corresponding structure, the terms were held indefinite — and the asserted claims invalid. That is the trap in full: a drafting choice made years earlier converts a functional claim element into an invalidity ruling. Which terms create exposure Generic, structure-light nouns paired with a function are the risk zone. Common examples include: module, unit, component, element mechanism, device, system (used generically) handler, processor, engine, controller (where no structure is specified) None of these is automatically fatal. The question is whether, read in context, the term conveys sufficiently definite structure to a person of ordinary skill in the art — and whether the specification backs it up. How to draft around it For software, algorithm, and AI-related patents: Disclose the algorithm, not just the outcome. Describe the actual steps, decision logic, and data transformations that perform each claimed function. A functional restatement of the claim is not an algorithm. Include supporting technical architecture. Execution flow, module interactions, data structures and handling rules, and the hardware environment where relevant. Recite structure in the claims where possible. Structural or structure-connoting language reduces the chance the presumption is overcome. Audit existing portfolios. Claims drafted before this line of cases hardened may carry exposure that only surfaces during litigation — the most expensive moment to discover it. The related drafting risks Functional claiming is one of several ways a patent’s own text narrows or destroys its scope during claim construction: Lexicography. If the specification expressly defines a term — particularly in a “Definitions” section, using phrases such as “refers to” or “unless otherwise specified” — courts will enforce that definition against the patentee. In Alnylam Pharmaceuticals v. Moderna (Fed. Cir. June 4, 2025), the patentee’s own definition of a claim term produced a narrow construction and a finding of non-infringement. Disclaimer. Statements in the specification or prosecution history such as “the invention must include,” “all embodiments include,” or “the essential feature of the invention is” can permanently narrow scope. Indefiniteness generally. Degree words such as “about,” “substantially,” “optimized,” or “high quality,” without accompanying numerical ranges or a defined test, can render a claim indefinite under the reasonable-certainty standard of Nautilus v. Biosig. FAQ Does a claim need the word “means” to trigger §112(f)? No. Where “means” is absent there is a rebuttable presumption against §112(f), but that presumption can be overcome if the term recites a function without sufficiently definite structure. What structure must a software patent specification disclose? Generally an algorithm — the specific steps performing the claimed function. Restating the function in the specification is not sufficient. What happens if the specification lacks that structure? The claim can be held indefinite and invalid, as the Federal Circuit affirmed in Fintiv v. PayPal (2025). General information only, not legal advice. Every case turns on its own facts.