Deploying Generative AI Without Losing Your IP: A Practitioner's Guide to Authorship, Disclosure, and Contracts
By Casey Scott McKay ·
This guide is the operational companion to the Marksy article on AI authorship: twelve numbered stages that take a company from "our designers started using Midjourney" to a defensible file. It covers building an AI use policy that actually preserves copyright in deliverables, designing a provenance record that will survive a deposition, deciding what to register and in what form, completing the Copyright Office disclaimer field by field with model text, and repairing applications already filed without disclosure through supplementary registration under 17 U.S.C. § 408(d). It then turns outward to contracts: a clause-by-clause redline of a model vendor agreement with ask, fallback, and walk-away positions on training rights, output ownership, indemnity scope, filter conditions, and IP warranties, plus the customer-facing representations you can and cannot give. Later stages cover auditing training data and provenance if you build or fine-tune a model, the trademark and right-of-publicity exposure in generated brands, voices, and personas, a day-by-day playbook for the first two weeks of an infringement claim over an output, and the recordkeeping that decides these cases in discovery. Every stage gives the governing authority, a realistic timeline and cost, and the specific mistake that costs money. A single worked example — Corvid Press, the graphic-novel imprint from the companion article — is carried from policy through registration to a claim letter.
IP and Technology > Copyright | Guide | Published 9 May 2026 - Updated 15 July 2026 | Casey Scott McKay - marksy.us
Summary. The doctrine is settled enough to work with: a machine cannot be an author, purely generated material is not protectable, and what survives is the human contribution. That is the companion article's job. This guide is the other half — what you actually do on Monday. Twelve stages take you from tool inventory to litigation file: writing an AI use policy that preserves copyright instead of merely prohibiting things, capturing provenance in a form that survives a deposition, choosing a registration path, completing the Copyright Office disclaimer field by field with model text, fixing applications you already filed wrong, negotiating a vendor agreement clause by clause with ask/fallback/walk-away positions, giving customers and clients disclosures that are true, auditing a training corpus if you build your own model, clearing generated brands and synthetic voices, running the first fourteen days of an infringement claim, and keeping the records that will decide the case. Each stage states the governing rule, the realistic cost and timeline, and the trap. Corvid Press — the imprint from the companion article — is carried through all twelve.
Keywords: ai use policy · generative ai governance · copyright office ai disclosure · limitation of claim · supplementary registration · human authorship record · prompt logging · provenance metadata · content credentials · vendor indemnity negotiation · output ownership clause · ai training rights carve-out · training data audit · corpus register · synthetic voice release · generated logo clearance · litigation hold · 17 u.s.c. 411(b) · eu ai act article 50 · ai contract clauses
This is premium Marksy content — the full document is available to subscribers.