Patent Risks of Using AI: Public Disclosure Concerns for Inventors

Why inventor interactions with external AI tools may jeopardize patent rights

Generative AI tools are quickly becoming part of the ordinary workflow for researchers, engineers, and product teams. Inventors may use AI to brainstorm alternatives, summarize references, refine technical explanations, draft slides, or prepare emails and reports. But when an inventor enters technical information into an external AI system, an important patent question arises: has the inventor just made a disclosure that could jeopardize patent rights? The answer is not always straightforward, but the risk is real and often underestimated. The issue is especially significant because U.S. law and non-U.S. law do not treat pre-filing disclosures the same way.

Under U.S. patent law, a claimed invention may be barred if, before the effective filing date, it was patented, described in a printed publication, in public use, on sale, or otherwise available to the public. U.S. law does include a limited one-year grace period for certain inventor-originated disclosures, but that grace period is not a complete safety net and should not be treated as a substitute for filing first. The United States Patent and Trademark Office’s (USPTO) discussion of AIA 35 U.S.C. § 102 likewise emphasizes that prior-art-triggering events include printed publications, public use, sale activity, and other availability to the public before the effective filing date.¹

Outside the United States, the consequences can be much harsher. The European Patent Office (EPO) applies the principle of absolute novelty: the state of the art includes everything made available to the public anywhere in the world by written or oral description, by use, or “in any other way,” before filing or priority. The EPO also makes clear that, except for narrow exceptions, an applicant’s own pre-filing disclosure can be cited against the application.² World Intellectual Property Organization (WIPO) materials likewise show that grace-period rules vary significantly by country and, in many jurisdictions, are limited, conditional, or unavailable.³

That is why AI use raises a patent issue. If an inventor types enabling technical details into an AI platform, the legal question is not merely whether the tool was “helpful”; it is whether the disclosure was made under conditions that preserve confidentiality, or instead under circumstances that could later be characterized as public or non-confidential. In its 2024 guidance on AI use before the USPTO, the Office specifically warned that AI tools can raise confidentiality concerns and noted that issues can arise from “sharing sensitive and confidential client information to third-party AI systems, including those potentially located outside of the United States.” The USPTO also reminded practitioners of their duty to preserve client confidentiality.⁴ Whether a particular AI interaction rises to the level of a patent-defeating disclosure will depend on the facts, including the applicable confidentiality terms, access controls, and how the system handles and retains user inputs.

Importantly, not all AI environments are the same. Some enterprise offerings are expressly designed to keep prompts and outputs within a protected commercial environment. For example, Microsoft states that for Microsoft 365 Copilot, prompts, responses, and data accessed through Microsoft Graph are not used to train the foundation large language models used by Copilot, and that the data remains within the Microsoft 365 service boundary. Microsoft also states that Microsoft 365 Copilot uses Azure OpenAI services, not OpenAI’s publicly available services.⁵ Anthropic states that, by default, it does not use inputs or outputs from its commercial products to train its models unless the customer affirmatively provides feedback or otherwise chooses to allow such use.⁶ OpenAI states in its enterprise privacy materials that, by default, it does not use business customer data to train its models.⁷

Those distinctions matter, but they do not eliminate the need for a patent-focused review. Even where a provider says business data is not used for model training by default, companies still need to evaluate the actual contract, service tier, settings, retention, logging, connector behavior, sharing permissions, and any optional feedback features. Microsoft, for example, notes that stored interaction data can include prompts and responses, that activity history is retained within the Microsoft 365 environment, and that administrators can manage or search that data through Microsoft Purview and related tools. Anthropic likewise notes that if users submit feedback, the related conversation may be stored and may be used for analysis and model training. In short, “enterprise” does not automatically mean “no patent risk”; it means the analysis becomes more fact specific and contract dependent.

The patent risk of AI use is not limited to disclosure. A second, separate issue is inventorship. The USPTO has made clear that only natural persons can be inventors on U.S. patent applications. USPTO guidance explains that patent protection may be available for AI-assisted inventions where one or more natural persons significantly contributed to the claimed invention, but AI itself cannot be named as an inventor. Thus, organizations should evaluate both (1) whether use of AI created a disclosure problem and (2) whether the inventorship record still accurately reflects human conception and contribution.⁸

Practical Takeaways for Inventors and Companies

The safest rule remains the old one: file before disclosure. That rule should now be extended to cover AI prompts, uploads, pasted technical text, schematic descriptions, experimental results, draft claims, and any other information that could reveal the invention in enabling detail. Because the EPO and many other systems are unforgiving as to pre-filing public disclosure, organizations pursuing international patent protection should assume that a problematic AI disclosure may impair foreign rights even if some argument might exist under U.S. grace-period law.

At a minimum, companies should consider implementing an internal AI policy specifically for inventors and R&D personnel. That policy should identify which AI tools are approved, whether only enterprise environments may be used, what technical information may or may not be entered, whether feedback features must be disabled, and when patent counsel must be consulted before using AI on invention-related material. The USPTO’s AI guidance underscores that confidentiality and accuracy issues are real when using AI tools in patent-related contexts, and provider documentation confirms that data handling varies materially by product and configuration.

Practical guardrails may include:

• No entry of unpublished invention details into public or consumer AI tools unless patent counsel has confirmed the risk is acceptable.
• Prefer approved enterprise environments with contractual data protections and clearly understood settings.
• Disable optional feedback features where feasible, and review logging/retention practices.
• Train inventors that “uploading to AI” is not the same as “thinking privately” and may have IP consequences.
• File early—often with a provisional application—before using AI to iterate on invention disclosures, abstracts, figure descriptions, or commercialization materials.
• Document human contribution to preserve inventorship positions for AI-assisted work.

Conclusion

AI can be a powerful productivity tool for inventors, but it also creates a new pathway for inadvertent disclosure. Whether a particular use constitutes a patent-defeating public disclosure will depend on the facts: what was disclosed, to whom, under what terms, with what confidentiality protections, and in which jurisdictions protection is sought. But the practical lesson is clear: inventors should treat external AI systems as potential disclosure channels unless the company has affirmatively vetted the tool, the contract, and the workflow for patent safety. In a world of rapid experimentation and global filing strategies, convenience today can become lost patent rights tomorrow.

References

1. USPTO, MPEP § 2152, Detailed Discussion of AIA 35 U.S.C. 102(a) and (b), https://www.uspto.gov/web/offices/pac/mpep/s2152.html.
2. European Patent Office, European Patent Guide, Chapter 3, 3.3 Novelty — Basic Principles, https://www.epo.org/en/legal/guide-epc/2024/ga_c3_3_1.html.
3. WIPO, Certain Aspects of National/Regional Patent Laws — Grace Period, https://www.wipo.int/export/sites/www/scp/en/national_laws/grace_period.pdf.
4. USPTO, Guidance on Use of Artificial Intelligence-Based Tools in Practice Before the United States Patent and Trademark Office, discussion of confidentiality and third-party AI systems, 89 Fed. Reg. 25609 (Apr. 11, 2024), https://www.federalregister.gov/d/2024-07629.
5. Microsoft, Data, Privacy, and Security for Microsoft 365 Copilot, Microsoft Learn (updated May 18, 2026), https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy.
6. Anthropic, Is my data used for model training? — Commercial Customers, Privacy Center (Mar. 16, 2026), https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training.
7. OpenAI, Enterprise privacy at OpenAI (updated Jan. 8, 2026), https://openai.com/enterprise-privacy/.
8. USPTO, Inventorship Guidance for AI-Assisted Inventions, 89 Fed. Reg. 10043 (Feb. 13, 2024), https://www.federalregister.gov/d/2024-02623.