Can you use ChatGPT for patents? What the USPTO says about AI and unfiled inventions

The Morse teamPublished October 29, 202610 min read

You can use ChatGPT for patents, but an unfiled invention typed into a personal ChatGPT account sits on OpenAI's servers, where authorized staff and service providers may read it and, unless you opt out, it may train OpenAI's models (OpenAI). As of October 29, 2026, no patent office we checked has said whether that counts as a public disclosure. The USPTO treats it as a confidentiality risk, and its guidance warns that AI systems "may retain the information that is entered by users" and may use it "to further train its AI models" (USPTO).

On October 29, 2026, we read the USPTO's April 11, 2024 Federal Register guidance on AI tools in full, its November 2025 inventorship guidance, 35 U.S.C. 102, the examining manual's sections on prior art and on examiners' internet searches, Articles 54 and 55 of the European Patent Convention with the EPO's 2026 examination guidelines, the EPO's and WIPO's own AI tools, and OpenAI's, Anthropic's, and Google's consumer pages. This isn't legal advice. Whether and how to patent an invention is a question for a patent attorney; this page covers who can read it once you type it.

Does using ChatGPT count as a public disclosure?

Nobody official has said. We found no statement from the USPTO, the EPO, or WIPO, and no court decision, on whether typing an invention into an AI chat makes it prior art. The closest we found is a university technology office: UT Southwestern wrote on February 25, 2025 that "there is no definitive word yet on this matter," while advising "strong caution" because OpenAI "is not obligated to keep confidential" what users upload (UT Southwestern). One law firm goes further. Marks & Clerk argued in January 2023 that ChatGPT input "is likely to be considered a public disclosure," since OpenAI's reviewers have no confidentiality agreement with the user (Marks & Clerk). No patent office or court has adopted that view.

If you plan to keep the invention as a trade secret instead, a parallel test applies: federal law protects it only if you've taken "reasonable measures to keep such information secret" (18 U.S.C. 1839).

How does US patent law decide what counts as public?

By whether the public could get at it. Under 35 U.S.C. 102(a)(1), an invention can't be patented if it was "patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date." The USPTO's examining manual calls that last phrase a "catch-all" that lets examiners focus on whether a disclosure was "available to the public," whatever form it took (MPEP 2152.02(e)). A document is a printed publication when "persons interested and ordinarily skilled in the subject matter or art, exercising reasonable diligence, can locate it." Internal documents stay out only under "an existing policy of confidentiality or agreement to remain confidential," and "mere intent to remain confidential is insufficient" (MPEP 2128).

No one can find a private chat by searching for it. The risk comes from what happens to the text afterward. The USPTO's guidance says that if confidential information is used to train AI, "that confidential information or some parts of it may filter into outputs from the AI system provided to others" (89 FR 25609). A chat you share by link is open to anyone who has the link, as our guide to AI chat share links shows app by app.

Does the one-year grace period cover an AI chat?

The US grace period covers disclosures that come from the inventor, and no statute or USPTO notice we read mentions AI chats. Section 102(b)(1) says a disclosure "made 1 year or less before the effective filing date" isn't prior art if "the disclosure was made by the inventor or joint inventor" or by someone who got the subject matter from them (35 U.S.C. 102). The manual calls these "grace period inventor-originated disclosures" (MPEP 2153.01(a)). It's a US rule, and Europe has no general grace period like it.

What about Europe's absolute novelty rule?

It's stricter. Article 54(2) of the European Patent Convention counts as prior art "everything made available to the public by means of a written or oral description, by use, or in any other way, before the date of filing" (EPO). Article 55 excuses a disclosure only within six months, and only for "an evident abuse in relation to the applicant or his legal predecessor" or a display at an officially recognised international exhibition (EPO). Your own chat with an AI fits neither.

The EPO's 2026 guidelines turn on confidentiality. Subject matter is public if "it was possible for members of the public to gain knowledge of it and there was no duty of confidentiality restricting the use or dissemination of this knowledge" (EPO Guidelines G-IV, 7.2.1). A secrecy agreement can be "express or tacit." The guidelines list "a relationship of good faith and trust" as a sign of a tacit one and "an ordinary commercial transaction" as a sign there was none (G-IV, 7.2.2). We found no Board of Appeal decision that says which of those a consumer AI subscription is.

What does the USPTO say about using AI for patent drafting?

It allows it, and it names confidentiality as the risk. The April 11, 2024 guidance says "there is no prohibition against using these computer tools in drafting documents for submission to the USPTO. Nor is there a general obligation to disclose to the USPTO the use of such tools," unless the use is material to patentability under 37 CFR 1.56(b) (89 FR 25609). Its section on confidentiality gives the inventor's case as its example. Disclosure "can happen, for example, when aspects of an invention are input into AI systems to perform prior art searches or generate drafts of specification, claims, or responses to Office actions." It ends: "before using these AI tools, it is imperative for practitioners to understand an AI tool's terms of use, privacy policies, and cybersecurity practices."

The confidentiality rule it cites, 37 CFR 11.106, binds patent attorneys and agents, not inventors. It requires "reasonable efforts to prevent the inadvertent or unauthorized disclosure of" client information, and the USPTO amended it in 2021 to align with ABA Model Rule 1.6, the confidentiality rule for lawyers. The signature rule, 37 CFR 11.18, "applies to anyone presenting a paper, including pro se applicants," so an inventor filing alone vouches for AI-drafted text as their own. If a patent attorney works on your invention, you can ask which AI tools they put it into.

The USPTO's later AI notice, the November 28, 2025 inventorship guidance, calls AI systems "instruments used by human inventors," "analogous to laboratory equipment, computer software, research databases." It says nothing new about confidentiality, and we found no later USPTO notice in the Federal Register that replaces the April 2024 guidance.

Can putting an invention into AI break export rules?

It can, the USPTO says. Its guidance warns that "AI tools may utilize servers located outside the United States, raising the likelihood that any data entered into such tools may be exported outside of the United States, potentially in violation of existing export administration and national security regulations or secrecy orders." Even on US servers, use "by non-U.S. persons may be deemed an export." A foreign filing license doesn't cover it: one "does not authorize the exporting of subject matter abroad for the preparation of patent applications to be filed in the United States" (89 FR 25609).

The consumer apps say where your text goes. OpenAI's help page says content "is stored on OpenAI systems and our trusted service providers' systems in the US and around the world" (OpenAI, read in an archive copy from October 14, 2026). Anthropic's privacy policy, effective September 10, 2026, says personal data is transferred "to our servers in the US, or to other countries outside the European Economic Area" (Anthropic).

How patent offices treat their own AI tools

The patent bodies' own AI tools and search rules show how they handle unpublished inventions. We read each of these on October 29, 2026.

The EPO built its classifier to help people pick classification codes from a description, and it still tells you to keep unpublished inventions out of it. WIPO built its assistant so queries run only on its own servers, the kind of answer the USPTO's guidance tells practitioners to look for in a tool's terms.

Do ChatGPT, Claude, and Gemini train on your invention description?

On a personal account, each can unless you change a setting, and each keeps some chats past deletion. OpenAI asks, "Please do not enter sensitive information that you would not want reviewed or used" (OpenAI), and Google asks you not to "enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services" (Google).

As of October 29, 2026, OpenAI may train on your ChatGPT chats unless you opt out, and a rated reply's "entire conversation" may still be used after you do (OpenAI). Anthropic keeps feedback and chats flagged for safety review "even if you opt out" of training (Anthropic). Google keeps Gemini chats that human reviewers saw for up to three years, and its reviewers include "trained reviewers from our service providers" (Google).

Each of those copies sits with a company under consumer terms, which is what the EPO's "duty of confidentiality" test and the USPTO's warning both ask about. The settings for each app are in our guides to whether ChatGPT trains on your data, Claude's privacy, and Gemini's privacy.

How Morse keeps an unfiled invention off AI company servers

Every test on this page turns on a copy of your invention sitting with a company that could read it, keep it, or train on it. That holds for the USPTO's warning that AI systems "may retain the information," for the EPO's "duty of confidentiality," and for the question of where the servers are. Morse is a web app with over 30 models, built so your chats aren't kept on its servers. Morse's servers don't keep messages or log message content, and in E2EE they relay text they can't read. You choose a privacy mode for each chat:

• E2EE, for the invention itself. Start the chat in E2EE before you describe how the thing works, and pick an open-weight model like GLM, Kimi, or DeepSeek. Your browser checks the secure enclave's hardware attestation, then encrypts the message so only that enclave can read it, and the reply is decrypted only in your browser. The enclaves are run by third-party providers, not by Morse, and the guarantee rests on the enclave hardware and attestation working as designed, as our guide to verifying an AI enclave's attestation explains. An E2EE chat can only continue in E2EE, so the rest of the conversation stays there too. E2EE chats come with Pro, at $19 a month.
• Private, for the background around it. It's the default for a new chat. Open-weight models like DeepSeek, Kimi, and GLM, plus Grok, run with zero data retention: Morse has zero data retention agreements with its inference, search, and other providers, so they keep nothing. Where a provider confirms zero retention on each reply, Morse checks it and refuses a reply that doesn't. Our explainer on zero data retention covers what that promise means.
• Anonymous, for general questions with GPT, Claude, or Gemini. Morse's proxy removes your account, email, and IP address and sends the message under Morse's own account, so the provider sees it without knowing it's you. It may keep the message for a limited time to check for abuse, under its own policy, so ask what a provisional application is here, not how your device works.

E2EE has no web search, so search for prior art the way USPTO examiners do for unpublished applications: in a separate chat, with terms limited to the general state of the art.

In every mode, your chat history is stored in your browser, encrypted, and unlocking it takes two keys: one that never leaves your device and one Morse's server gives you when you sign in. Morse doesn't train AI on your content, sell your data, or show ads, so there's no training setting to find (Morse privacy policy). Clearing your browser or deleting your account deletes your chats, and Morse can't recover them, so keep your own dated record of the invention somewhere else.

Two questions stay with your counsel whatever tool you use. For export-controlled technology or an invention under a secrecy order, the USPTO asks where a tool's servers are and who can reach them, so settle that before you type it into any AI tool, Morse included. And no patent office has ruled on any kind of AI chat, so if a filing is close, ask your patent attorney first.

Try a chat from the Morse home page without an account, or start free and move to Pro when the invention itself needs E2EE.