Does the United States Patent Office Protect Unpublished Applications When Searching the Internet and Artificial Intelligence Databases for Prior Art?

Short answer: Yes, but…

Many practitioners in sensitive technology areas file patent applications with non-publication requests or may abandon their applications if examination is not going well to avoid public disclosure of their application. Whether it is patent subject matter eligibility issues or novelty/obviousness issues that lead a practitioner to turn a patent into a trade secret, some applications may require some secrecy unless a patent will be granted. However, do the U.S. Patent and Trademark Office (USPTO) policies and rules ensure inadvertent outing of the invention during searching? And do they continue to do so now that the Office is using artificial intelligence databases to search? They are trying to.

The USPTO outlines several policies on internet usage for prior art searching. For example, MPEP 904.02(c) discusses the Internet Usage Policy memorialized in 64 F.R. 33056 (June 21, 1999). Article 9 of the Policy states in relevant part, “When the Internet is used to search, browse, or retrieve information relating to a patent application, other than a reissue application or reexamination proceeding, Patent Organization users MUST restrict search queries to the general state of the art. Internet search, browse, or retrieval activities that could disclose proprietary information directed to a specific application, other than a reissue application or reexamination proceeding, are NOT permitted.”

To help police Article 9, Article 10 requires “All Patent Organization users of the Internet for patent application searches shall document their search strategies in accordance with established practices and procedures as set forth in MPEP 719.05 subsection I.(F).”

MPEP 904.02(c) goes on to discuss the limitations on using social media for prior art searching. For example, it states in relevant part, “All use of the Internet by examiners must be conducted in a manner that ensures compliance with confidentiality requirements in the statutes, including 35 U.S.C. § 122, and regulations. As part of an Internet search, examiners may visit social media websites that provide for public interactions, but are not authorized to participate on these websites or otherwise solicit assistance with patent examination even for published applications. For example, an examiner is not authorized to request users of a website to provide additional information not found on the website, elaborate further on information already posted on the website, or post new information on the website. Additionally, any Internet searching that is conducted is to be limited to searching for the information necessary for examination of the application or proceeding, such as the state of the art or the presence or absence of technical features in the prior art.”

In recent years, the USPTO has made artificial intelligence (AI) databases available to Examiners for use in prior art searching. For example, the USPTO’s internal Patents End-to-End (PE2E) search suite contains an AI-based “similarity search” feature. The similarity search feature receives examiner-selected application information, including the specification, as input and uses trained AI models to output a list of domestic and foreign patent documents that are similar to a patent application being searched. While the AI-based similarity search feature receives sensitive data, such as an application’s full specification, the USPTO implements security controls to maintain patent application confidentiality as required by 35. U.S.C. § 122(a). For example, all search queries are run within USPTO system boundaries using a Federal Risk and Authorization Management Program authorized cloud. The confidentiality provision of 35. U.S.C. § 122(a) also prevents text snippets used in a similarity search from being listed in the PE2E search history for certain applications.

But what happens when Examiners use external AI websites to search for prior art? The USPTO website provides a list of commercial databases Examiners can use to search for prior art, including the “IP.com| InnovationQ Plus | Patent search tool that features fully discoverable content from IEEE, including full-text journals and conference papers.”

The InnovationQ Website states it uses AI to build forward searching using past searching, including:

• “SEMANTIC GIST®: Harness the power of our Semantic Gist® engine, which aggregates vast amounts of global IP data, encompassing patent and non-patent literature. The comprehensive library of patent and non-patent literature is supplemented with industry-exclusive access to IEEE and OnePetro content. Accurate, advanced, and actionable results are at your fingertips.”
• “SYNOPSIS AI: Revolutionize IP evaluation with Synopsis AI. This pioneering approach swiftly distills essential IP content, transforming pages into succinct paragraphs for expedited analysis and faster turnarounds.”
• “SEMANTIC MAP: Navigate the competitive landscape effortlessly with the Semantic Map (concept map). Visualize relationships across multiple vectors, spotlighting key terms to streamline searches. Identify unexplored opportunities by pinpointing white space.”
• “HD-ANALYZE: Immerse yourself in HD-Analyze’s integrated data visualization. Drastically reduce search-related time and costs while gaining crucial insights. Our tool empowers you to make informed decisions efficiently.”

The InnovationQ+ system allows for prior art searching using text inserts from patent applications. Thus, sensitive application data may be exposed to a third-party AI system. While data from unpublished applications are unlikely to appear in future InnovationQ+ searches, this data may be used for internal purposes, such as training AI models.

Knowing this, patent practitioners need to pay close attention to their Examiner search strategy forms submitted with an Office Action or Notice of Allowance to stay informed on where their invention information may have been used when the Examiner performed their prior art searching duties. Moreover, for inventions with borderline non-patentable subject matter or in highly crowded fields, applicants need to determine whether they want their technology exposed to an AI database during searching.