Skip links
AI and patent rights

AI and Patent Rights: Can Uploading an Invention Destroy Them?

Uploading a patent draft, drawings, or confidential invention information to a consumer AI platform may create unexpected patent risks. We examine § 102, Sanho, Helsinn, Celanese, and what inventors should do before clicking “Upload.”

Can Uploading an Invention to ChatGPT or Another AI Platform Destroy Patent Rights?

Questions about AI and patent rights are becoming increasingly important as inventors use artificial intelligence to review invention disclosures, draft patent applications, drawings, test data, and other confidential technical information. If they are not careful, they may jeopardize valuable patent rights before an application is ever filed. Under 35 U.S.C. § 102, certain pre-filing disclosures can create a statutory bar and prevent an inventor from obtaining a patent.

AI and patent rights
Before uploading your invention into an AI platform, learn about the potential risks.

When an inventor clicks “Upload” and sends information about an invention to an AI platform, has the inventor just lost patent rights?

It is 10:47 on a Tuesday night. An engineer at a small Midwestern manufacturer is working on something the company has kept closely guarded for months.

The new design could cut production costs by nearly a third, substantially increase the company’s margins, and change its position in the market. If it works as expected, it could give the company a significant advantage over its competitors and help secure business with a Fortune Global 50 customer. Only a handful of people inside the company know about the new design, and even fewer have seen it.

The prototype works. The engineering drawings are finished. The test results are promising. A draft patent application prepared by the company’s patent attorney is sitting on the engineer’s computer, but it has not yet been filed.

She wants one more set of eyes on it.

Instead of waiting until morning to call the patent attorney, she opens an AI application such as ChatGPT, Claude, or Gemini. She uploads the drawings and draft patent application, along with photographs of the new design.

Then she types:

“Review this invention. Tell me what I am missing, identify any weaknesses, and suggest ways to design around it.”

Within seconds, the AI platform provides its analysis.

The engineer closes her laptop and goes to bed.

But something important has happened. The new design, a company secret that had only been known inside the business has now been disseminated outside the company. The drawings, photographs, and invention disclosure contain everything needed to understand and practice the invention.  That information is now available to the developers of the AI platform.

Has the company just lost its patent rights?  Maybe.

Can Using AI Affect Patent Rights Under § 102?

Public disclosure is part of the patent bargain.  As a result, patent law has a long history of addressing disclosures, publications, public uses, and prior sales. AI looks at those familiar questions from a slighltly different lens.

Under 35 U.S.C. § 102(a)(1), a person is not entitled to a patent if, before the effective filing date, the claimed invention was:

“patented, described in a printed publication, or in public use, on sale, or otherwise available to the public.”

The statute does not say that every time an inventor shares an invention with another person or company, patent rights are automatically lost. Nor does it contain a special rule for ChatGPT, Claude, Gemini, or another AI platform.

The question is what happened to the invention when it was uploaded, and whether what happened falls within § 102.

Uploading the Invention Raises the Risk

There is an important difference between asking an AI platform:

“How could I make a more efficient pump?”

and uploading drawings showing a new pump, photographs of the prototype, test results, and a invention disclosure explaining exactly how the invention works and what makes it different.

The first may disclose little or nothing about the inventor’s actual invention. The second may disclose nearly everything.

That can matter for at least two reasons.

First, if the uploaded documents describe every important part of the later-claimed invention, it becomes much easier to establish exactly what was disclosed and when.

Second, the documents may provide evidence that the invention was ready for patenting.

In Pfaff v. Wells Electronics, Inc., 525 U.S. 55 (1998), the Supreme Court held that an invention may be ready for patenting when it has been reduced to practice or when the inventor has prepared drawings or other descriptions sufficiently specific to enable a person skilled in the art to practice the invention. The Supreme Court later confirmed in Helsinn Healthcare S.A. v. Teva Pharmaceuticals USA, Inc., 586 U.S. 123 (2019), that Congress did not discard the established meaning of the on-sale bar when it enacted the America Invents Act.

Engineering drawings and a invention disclosure may therefore be used as evidence of a statutory-bar if there is a later patent infringement dispute.

But evidence that an invention was ready for patenting is not itself a statutory bar. Something else must still happen.

Sanho: A Private Disclosure Is Not Necessarily a Public Disclosure

The Federal Circuit’s decision in Sanho Corp. v. Kaijet Technology International Ltd., 108 F.4th 1376 (Fed. Cir. 2024), approached public disclosure from an unusual direction.

The issue was whether a private sale meant that the invention had been “publicly disclosed” under § 102(b)(2)(B).

The Federal Circuit said no.

Kaijet had challenged Sanho’s patent in an inter partes review. The Patent Trial and Appeal Board found several claims obvious based on combinations that included an earlier-filed patent application known as Kuo. Because Kuo’s effective filing date preceded Sanho’s patent filing, Kuo qualified as a prior art reference under § 102(a)(2).

Sanho tried to take Kuo out of the prior art picture.

Section 102(b)(2)(B) provides an exception to a prior art being used to invalidate a patent when the later-filed patent had been “publicly disclosed by the inventor” before the reference’s effective filing date.

Before Kuo was filed, Kuo disclosed its invention to Sanho.  Kuo provided Sanho with a sample and offered to sale 15,000 units to Sanho.  According to Sanho, this occured more than a month prior to the effective filing date of the Kuo patent.  According to Sanho, there was a public dislosure of the Kuo invention and thus, Kuo was prevented from being used as prior art.

The Federal Circuit disagreed.

The court held that Congress’s use of “publicly disclosed” in § 102(b)(2)(B) meant something more than simply “disclosed.” The inventor had completed a private, nonconfidential commercial transaction, but there was no evidence that the sale was publicized or that the relevant features of the invention had been taught to anyone beyond Sanho. The invention had not been shown to have been made available to the public.

As a result, Sanho could not invoke the § 102(b)(2)(B) exception. Kuo remained prior art, and the challenged claims remained unpatentable.

Of note, Sanho did not hold that a private disclosure can never create prior art. Nor did it decide whether uploading an invention to an AI platform makes the invention “otherwise available to the public” under § 102(a)(1).

It decided a more limited question: whether the inventor’s earlier private sale was sufficiently public to give the inventor the benefit of the § 102(b)(2)(B) exception.

The answer was no.

Still, the court’s reasoning offers an relevant lesson for AI use:

Sending an invention outside the company is not necessarily the same thing as making the invention available to the public.

AI Use Is Not the Same as a Publicly Available

Consider our hypothetical engineer again.

She uploaded the company’s invention disclosure and drawings into an AI platform. Assume the files can be accessed by the company operating the platform but cannot be searched for or viewed by other users.

Has the invention left the company’s possession? Yes, but has it been publicly disclosed?

Sanho suggests the answer is not automatic.

In Sanho the inventor assumed that any outside disclosure was a qualifying disclosure. However, a private disclosure may not provide the types of disclosure which Congress established as part of the incentive for inventors to actually file patent applicaitons disclosing their inventions to the public.

If an independent third party later files a patent application covering the same invention, a private AI upload would not necessarily allow the original inventor to invoke § 102(b)(2)(B) against that later patent filing. Sanho suggests that the statutory exception requires an actual public disclosure, not merely an upload into an AI platform.

There are different protections for a later applicant who obtained the invention directly or indirectly from the inventor. Section 102(b)(2)(A) addresses that situation.

The fact that a document was sent over the Internet into an AI platform does not necessarily mean that the invention described in it was available to the public.

AI Platform Terms and Settings Matter

Inventors should not assume that information uploaded to an AI platform is confidential.

Most consumer AI platforms reserve some right to use what users submit, although the scope of that use often depends on the user’s account and settings.

For individual ChatGPT accounts, OpenAI’s current terms permit it to use content to provide, maintain, develop, and improve its services. Users can opt out of having new content used to train its models. OpenAI also states that uploaded files may be used to improve its models when the applicable setting is enabled.

Anthropic allows users of consumer Claude accounts to choose whether their chats and related content may be used to improve Claude. Turning that setting off generally prevents new chats from being used for future model training, subject to stated safety-related exceptions.

Google states that when Gemini’s “Keep Activity” setting is on, chats and materials shared with Gemini, including files and photographs, may be used to improve Google services and train generative AI models. Users can turn that setting off for future chats.

None of these terms, standing alone, tells us whether an invention has become prior art. But they may help establish what happened to the invention after it was uploaded.

In a later patent dispute, relevant questions could include:

  • What did the inventor upload?
  • What did the platform’s terms allow the company to do with it?
  • Was AI training enabled?
  • Could employees, contractors, or others review the material?
  • Could another user access the information?
  • Was the invention ever disclosed beyond the AI company?

These are simple factual questions. Their answers could determine whether an inventor merely disclosed information to a third party or whether the invention was actually made available to the public.

Does Free AI Access Put the Invention “On Sale”?

Most consumer AI platforms offer free versions. Suppose an inventor uses one of those services and permits the platform to use uploaded information for AI training. One could argue that something of value has been exchanged: the inventor receives access to the service, while the AI company receives rights to use the inventor’s information.

Why does that matter?

Section 102(a)(1) bars a patent when, before the effective filing date, the claimed invention was “on sale.” Under Pfaff v. Wells Electronics, Inc., 525 U.S. 55 (1998), the on-sale bar requires a commercial offer for sale and an invention that is ready for patenting. The Federal Circuit has further held that the offer must be sufficiently definite to form a binding contract upon acceptance, assuming consideration. Group One, Ltd. v. Hallmark Cards, Inc., 254 F.3d 1041, 1048 (Fed. Cir. 2001).

An exchange of value may therefore matter. But it is not enough. The claimed invention itself must be the subject of the sale or offer.

In The Medicines Company v. Hospira, Inc., 827 F.3d 1363 (Fed. Cir. 2016) (en banc), the patent owner paid a manufacturer to produce batches of a patented pharmaceutical product. The Federal Circuit held that the transaction was a purchase of manufacturing services, not a sale of the patented invention. Commercial benefit alone, the court explained, does not trigger the on-sale bar.

In re Kollar, 286 F.3d 1326 (Fed. Cir. 2002), provides an even closer analogy. There, the inventor disclosed technical information about a claimed process and granted another company rights to commercialize it in exchange for royalties and other rights. The court held that licensing the process and providing the “know-how” needed to practice it did not place the invention on sale.

That distinction may apply to AI. Even if an AI platform receives something of value from the right to use uploaded information for training, the transaction may concern information about the invention, rather than a sale of the invention itself.

There can be harder cases. If the claimed invention is a software or process invention and the AI platform actually performs the claimed process as part of the service, the analysis could change. For example, in Scaltech, Inc. v. Retec/Tetra, LLC, 269 F.3d 1321 (Fed. Cir. 2001), the Federal Circuit held that commercial offers to perform a claimed waste-treatment process for customers triggered the on-sale bar.

Secrecy does not necessarily avoid that result. In Helsinn Healthcare S.A. v. Teva Pharmaceuticals USA, Inc., 586 U.S. 123 (2019), the Supreme Court held that a qualifying commercial sale can trigger the AIA on-sale bar even when details of the invention remain confidential. Likewise, in Celanese International Corp. v. International Trade Commission, 111 F.4th 1338 (Fed. Cir. 2024), the Federal Circuit held that sales of products made using a secret manufacturing process could bar later patent claims directed to that process.

This remains an unsettled area of patent law. Free AI access, even when the platform may use uploaded information for training, does not by itself appear to place an invention “on sale.” The more important question is whether the claimed invention itself was commercially sold, offered for sale, or, for a process claim, whether performance of the claimed process was commercially offered.

Likewise, asking whether an AI upload was “private” answers only part of the § 102 question.

So, Did the Engineer Lose the Patent?

Return to Tuesday night at 10:47.

Did uploading the company’s secret invention to an AI platform destroy its patent rights?

Not necessarily. But it created a risk that did not need to exist.

There does not appear to be a reported Federal Circuit decision holding that uploading an invention to an AI platform, by itself, makes the invention “otherwise available to the public” under § 102(a)(1).

But that does not make the upload harmless.

Sanho shows that a private disclosure is not necessarily a public disclosure. Helsinn and Celanese show that secrecy does not prevent a separate statutory bar when the conduct independently qualifies as an on-sale event.

And when an inventor uploads a patent draft, drawings, test results, photographs, source code, or other detailed technical material, the upload may create clear evidence of what the invention was, when it existed, and what information left the company.

The practical advice is therefore straightforward:

File before uploading an invention to a consumer AI platform whenever possible.

Businesses should also establish clear rules governing the use of AI with unfiled inventions and confidential technical information. Inventors should know what they are uploading, what the platform’s terms permit, and whether their information may be retained, reviewed, or used for training.

If an invention has already been uploaded, do not assume the patent rights are lost. Document the upload, the materials and date of the upload, the account settings, and the AI terms that applied at the time. Contact one of our patent attorneys to evaluate whether the disclosure raises an issue under § 102, whether the one-year U.S. grace period may apply, and whether foreign patent rights may be affected.

For now, the safest rule remains the simplest:

If the invention matters, file first. Use AI later.

Questions About AI and Patent Rights?

If you are developing a new invention, have questions about using AI during the patent process, or have already uploaded confidential invention information to an AI platform, the patent attorneys at IPCenter can help evaluate the risks and help provide an appropriate filing strategy.

Contact one of our patent attorneys before a preventable disclosure becomes a patent problem.

This article is for general informational purposes and does not constitute legal advice. AI platform terms, privacy settings, and data-use policies change frequently. The discussion of ChatGPT, Claude, and Gemini reflects publicly available information reviewed in August 2026.