Navigating Disclosure and Confidentiality in AI-Generated Intellectual Property Across U.S. IP Law

Generative artificial intelligence is rapidly reshaping how companies develop creative assets such as copy, logos, designs, and invention disclosures. However, the critical legal question is not merely about privacy but centers on whether and to what extent intellectual property owners must disclose AI’s role in creating these assets.

In the United States, there is no uniform confidentiality rule for AI-assisted creations. Instead, each branch of intellectual property law—copyright, patent, and trademark—draws its own boundary between compelled disclosure and process secrecy. These boundaries align with the core doctrines underpinning each regime: human authorship in copyright, human inventorship in patent law, and source identification in trademark law.

The constitutional foundation explains this divergence. Article I, Section 8, Clause 8 of the U.S. Constitution empowers Congress to secure rights for “Authors” and “Inventors,” which is why copyright and patent law remain focused on whether a natural person contributed the legally relevant creative or inventive input. Landmark Supreme Court cases such as Burrow-Giles Lithographic Co. v. Sarony and Feist Publications, Inc. v. Rural Telephone Service Co. have emphasized that copyright requires originality stemming from a human author rather than a purely mechanical process.

Trademark law, by contrast, is structured differently. The operative statutes—15 U.S.C. §§ 1051 and 1052—along with USPTO application regulations, focus on whether a designation is distinctive, non-confusing, and used in commerce. Supreme Court rulings like Two Pesos, Inc. v. Taco Cabana, Inc. and Jack Daniel’s Properties, Inc. v. VIP Products LLC have characterized trademark law in terms of source identification rather than authorship.

Under copyright law, the statutory starting point is 17 U.S.C. § 102, which protects original works of authorship, and 17 U.S.C. § 201, which vests ownership initially in the author. The U.S. Copyright Office’s 2023 registration guidance explicitly requires applicants to disclose AI-generated content included in a work submitted for registration. Applicants must identify human-authored material in the “Author Created” field and exclude AI-generated content exceeding a de minimis threshold in the “Material Excluded” section.

Importantly, the Copyright Office instructs applicants not to name AI systems or vendors as authors or co-authors, since claims must be limited to human authorship. The office’s forthcoming 2025 Part 2 report on copyrightability reinforces this stance, concluding that prompts alone do not demonstrate sufficient human control over expressive elements, whereas creative selection, arrangement, and meaningful modification of AI-generated material may support protection for the human-authored aspects of a final work.

This position aligns with appellate authority. In Thaler v. Perlmutter, the D.C. Circuit held that the Copyright Act requires works to be authored initially by a human being, while clarifying that human authorship can include works created with AI assistance if the legally relevant contribution is human.

From a confidentiality perspective, copyright is the least private of the three IP systems. Registration records become publicly accessible, and the Copyright Office’s privacy FAQ and public records portal confirm that registration files are open for public inspection. However, the disclosure requirement is categorical rather than operational: current guidance does not compel applicants to reveal AI prompts, model settings, training data, or the complete workflow behind the work.

Patent law begins with 35 U.S.C. § 100(f), defining an inventor as an “individual,” and 35 U.S.C. § 115, which requires an inventor’s oath or declaration. The Federal Circuit’s decision in Thaler v. Vidal held that the Patent Act requires inventors to be natural persons.

The USPTO’s Revised Inventorship Guidance for AI-Assisted Inventions clarifies that no separate inventorship test exists for AI. The same legal standard applies to all inventions, treating AI systems as tools used by human inventors. The USPTO generally presumes the inventors named on the application data sheet or oath are the actual inventors unless evidence suggests otherwise.

This distinction is crucial for disclosure. Current patent statutes and guidance focus on who conceived the claimed invention rather than mandating disclosure of every software tool involved in the inventive process. Thus, patent law does not impose a general requirement to list AI models, prompts, or vendors merely because AI was used during development.

However, patent law does require human responsibility for filings. The USPTO’s 2024 AI tools practice guidance reminds practitioners that existing rules apply when AI is used before the office, including the duty of reasonable inquiry and ensuring filings have evidentiary and legal support. The guidance also highlights client confidentiality concerns, warning that using external AI systems for drafting or prior art searching may expose sensitive client information to third parties.

Patent practice maintains a meaningful zone of prepublication secrecy. Under 35 U.S.C. § 122 and 37 C.F.R. § 1.14, pending patent applications are generally kept confidential unless publication rules mandate disclosure. Therefore, the patent system offers more confidentiality at the front end than copyright registration.

Trademark law does not inquire into who authored a mark, logo, or brand name. Instead, it focuses on ownership, goods or services covered, filing basis, and whether the designation functions as a source identifier. Current USPTO rules in 37 C.F.R. § 2.32 specify required contents of trademark applications but do not require applicants to disclose how the mark was conceived or whether generative AI assisted in its creation.

Consequently, AI-assisted brand names or visual brand indicia may be registrable if they are distinctive and properly used in commerce. The disclosure question is largely absent because trademark validity does not depend on human authorship or conception in the way copyright and patent law do.

A practical limitation exists: if an application is filed on a use basis, the applicant must provide a genuine specimen showing actual marketplace use. The USPTO’s specimen guidance emphasizes that specimens must reflect what consumers truly see in commerce. AI-generated mockups or idealized marketing images may pose evidentiary challenges if they do not demonstrate real use, even though AI provenance itself is not a disclosure item.