Overview
On June 10, a startup called Fearn announced a $5.5 million seed round and a product: upload your invention notes, get a complete patent application back, $2,000 flat. Business Insider called it the TurboTax for patents. Fearn did not file the application for you. Its home page that day told inventors to take the draft to their own lawyers and “File instantly with your trusted attorneys, and never miss your priority filing date.”
Six weeks later, Fearn had stopped sending inventors elsewhere. It now offers prosecution in-house, through Fearn Legal, LLC, which it describes as “a patent prosecution firm, licensed to file and prosecute patents across the USPTO.” Fearn files the application and carries it through examination on a staged fee schedule. If the examiner allows no claims, it refunds the cost of the non-provisional filing.
The USPTO’s 2024 guidance on AI tools is permissive about the drafting itself: “there is no prohibition against using these computer tools in drafting documents for submission to the USPTO.” There is also no general obligation to tell the Office you used them.
What the guidance will not give up is the human signer, a requirement that predates the software. Nearly all correspondence filed at the Patent Office must bear the signature of a natural person, and the rule exists to “ensure that documents drafted with the assistance of AI systems have been reviewed by a person and that person believes everything in the document is true and not submitted for an improper purpose.” The guidance addresses the machine directly: “It would not be acceptable for the correspondence to have the signature of an AI tool or other non-natural person.” An AI also cannot hold the USPTO account needed to file.
Under 37 C.F.R. § 11.18(b), whoever presents a paper to the Office certifies that it rests on “an inquiry reasonable under the circumstances.” The guidance is blunt about what will not qualify: “Simply relying on the accuracy of an AI tool is not a reasonable inquiry.” The certification binds “whether a practitioner or non-practitioner,” so it reaches a pro se applicant. The inventor who files his own AI-drafted application takes it on personally.
The USPTO keeps a public register of the practitioners it admits. A search by firm name returns one practitioner at Fearn Legal. Samuel Redford is a patent agent rather than an attorney, which the rules allow. Fearn lists him as its Patent Partner and publishes his registration number.
The reasonable inquiry under 37 C.F.R. § 11.18(b) has real work to do. The 2024 guidance warns that an AI drafting a specification may introduce “alternative embodiments which the inventor(s) did not conceive and applicant seeks to patent.” An embodiment nobody actually invented raises two questions at once: whether the named inventors conceived what is claimed, and whether the specification supports it under § 112. The file wrapper need not disclose that AI drafted the application, so the record may say nothing about where a given embodiment came from.
The inventorship question carries a duty of its own. Under 37 C.F.R. § 1.56, those preparing and prosecuting an application, including each named inventor, must disclose information material to patentability. The guidance says that once inventorship is in question, information about the interaction with the AI system “could be material and, if so, should be submitted to the USPTO.” The duty “cannot be transferred to another person or a computer system such as an AI tool.”
So the question about any AI-drafted patent is who made the reasonable inquiry before signing it. The register will name the practitioners. The question itself could arrive years later—from opposing counsel, in a deposition about how the application was written.
The Weekly Inference is a recurring Step Into IP feature. Nothing here is legal advice.
