Most inventors and business executives know that to obtain a patent, the claims of the application must be novel and non-obvious over the prior art. However, the scope of “prior art” is broader than most people realize. According to the statute, 35 U.S.C. §102(a)(1), “prior art” includes anything that was “patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the invention”.
Unlike trade secret misappropriation, which requires actual copying of something, prior art does not need to have been accessed by anyone to block a patent application or invalidate a patent. The only requirement is that it is available to the public. This includes information made public by the inventor or applicant themselves. While most cited prior art comes from patents, published patent applications, and scientific papers, not all references are so formal. The authors have seen web pages, online product listings, and even YouTube videos cited against applications during prosecution. Prior art need not be available online or even in your country. For example, an otherwise-unpublished student thesis on a shelf in a university library is sufficiently available to the public to count as prior art. As long as the library is accessible and an interested party could find it on the shelf, the contents of the thesis are considered publicly available. All’s fair in love, war, and patent law.
With the advent of publicly available AI tools comes a potential trap for inventors and applicants. Absent contractual provisions, prompts submitted to a publicly available AI model, such as ChatGPT, Claude, Gemini, etc., may become training data for future versions of the AI model. Therefore, if an inventor describes their invention in an AI prompt, that invention may be added to that AI model’s knowledge banks and may be accessed to answer future prompts from anyone. Unwary inventors who have submitted descriptions of their inventions to public AI models, e.g., to further refine their idea, or to draft a patent application themselves, may eventually find their patent application rejected or their patent invalidated on the basis that their invention was available to the public, regardless of whether anyone saw it or not.
Due to the relative infancy of generative AI models, and the slow-moving nature of patent prosecution and examination, there is not yet any caselaw deciding whether a prompt to a publicly available AI model constitutes a public disclosure. Sooner or later, though, (and I highly expect sooner) a judgment on this question will be issued, and if the court says “yes”, many patents may suddenly become invalid.
Here is what we recommend:
- If you or one of your inventors has already submitted information about an invention into a publicly available AI model, you should find out the date that prompt was submitted. In the U.S., a one-year grace period exempts prior art created by an inventor or someone who obtained the information from an inventor. Contact a patent attorney and make sure that a patent application is filed within one year of the submission of the prompt. Most countries outside of the U.S. do not have this grace period, unfortunately, although the scope of prior art can vary across jurisdictions.
- If you or one of your inventors is considering using AI to refine an invention or to draft a manuscript, a patent application, or an invention disclosure, then avoid using a publicly available AI model and instead use a private AI model with a contractual guarantee that inputs to the model will not be used as training data. ChatGPT, Claude, and Gemini, for example, allow you to opt out of having your inputs used as training data.
If you have any questions regarding this alert, please reach out to one of the authors.
