{"id":114136,"date":"2026-05-19T14:17:02","date_gmt":"2026-05-19T08:47:02","guid":{"rendered":"https:\/\/greyb.com\/?p=114136"},"modified":"2026-05-29T17:08:28","modified_gmt":"2026-05-29T11:38:28","slug":"ai-sep-invalidation","status":"publish","type":"post","link":"https:\/\/greyb.com\/blog\/ai-sep-invalidation\/","title":{"rendered":"6 AI Tools, One Sequence: GreyB’s Six-Layer SEP Patent Invalidation Process"},"content":{"rendered":"\n

Of the 800-plus SEP invalidation searches GreyB completed in the past year, the decisive prior art came from outside patent databases in most cases.<\/p>\n\n\n\n

The key evidence often did not come from a granted patent or a published research paper. It came from a 3GPP working group proposal, a draft contribution, an IEEE meeting document, or an email thread written before the final standard was published.<\/p>\n\n\n\n

That matters because most search workflows are not built to handle this type of material. Standard document databases and SSO archives are built for document retrieval. Boolean search is constrained, cross-source filtering is absent, and AI-assisted retrieval is unavailable in most SSO systems. Analysts often open documents one by one, follow trails that lead nowhere, and work through tens of thousands of pages while facing litigation holds, IPR filing deadlines, or FRAND negotiations<\/a>.<\/p>\n\n\n\n

\"image\"<\/figure>\n\n\n\n

The results from those 800-plus searches show a clear pattern. In 70% of cases, the search produced strong results. More than 60% of those searches produced Category X, 1, or 2 references. These references can directly challenge novelty and support strong invalidation arguments. The remaining 40% produced Category Y references and combinations that supported obviousness arguments.<\/p>\n\n\n\n

\"image\"<\/figure>\n\n\n\n

Those results did not come from search expertise alone. They came from six AI tools operating alongside analyst judgment, each used to remove friction at a specific stage of SEP prior art search where manual-only workflows lose time, miss sources, or stop short of the most consequential evidence.<\/p>\n\n\n\n

What follows is a description of each tool, what it does, and where it changed the outcome of a real SEP patent invalidation project.<\/p>\n\n\n\n

The Six-Layer Search Architecture Behind GreyB\u2019s SEP Invalidity Work<\/h2>\n\n\n\n

SEP invalidity searches fail when the standard is treated as a document repository rather than a development record. The analyst first needs to reduce the asserted claim to the exact technical feature it covers. That feature then needs to be mapped to the standard, traced through older drafts and meeting contributions, and tested against the patent\u2019s priority date.<\/p>\n\n\n\n

GreyB\u2019s workflow follows that sequence.<\/p>\n\n\n\n

The process combines analyst judgment with six internal systems that reduce the most failure-prone parts of the SEP prior art search. These systems support claim interpretation, essentiality confirmation, standard backtracking, meeting document retrieval, email archive search, technology evolution mapping, and clustered patent search.<\/p>\n\n\n\n

\"image\"<\/figure>\n\n\n\n

Each layer narrows the search field and opens more precise search paths.<\/p>\n\n\n\n

The tools do not replace SEP search expertise. They remove friction from the work that slows analysts down, including reading dense claims, locating standard sections, searching thousands of proposals, tracing working group discussions, and building a timeline of how the claimed feature evolved.<\/p>\n\n\n

Also Read:<\/h4>
\"AI<\/div>
AI Just Made SEP Claim Mapping 80% Faster – 600 Hours of Human Work to 14 Days<\/a><\/div>Read Post <\/i><\/a><\/div><\/div><\/div>\n\n\n\n

The analyst still decides which references matter, how they map to claim elements, and whether they support anticipation, obviousness, or a stronger invalidity position.<\/p>\n\n\n\n

Strong SEP prior art rarely comes from a single keyword query. It usually appears through a sequence of steps. The analyst needs to understand the claim, confirm overlap with the standard, identify the technical feature, trace the feature backward, locate the earliest public disclosure, and test that disclosure against the priority date.<\/p>\n\n\n\n

1. The Hour Lost Before Search Begins and How to Reclaim It<\/h2>\n\n\n\n

The first goal in any SEP prior art search is building a precise understanding of the claim. This sounds straightforward, but it is where analysts lose the most time and where many interpretation errors begin.<\/p>\n\n\n\n

SEP claims are technically dense. The underlying technology is often complex, and the claim language is compressed. The entities, conditions, and interactions are difficult to visualize from text alone. An analyst working only from written claim language starts at a disadvantage. <\/p>\n\n\n\n

The Advanced Patent Understanding Tool addresses this directly. Input the patent number, and the tool generates a structured breakdown of the invention, the problem it addresses, the proposed solution, and the narrowest defensible reading of each claim element. Alongside this, the tool renders the claim as a visual diagram, making communication flows, entities, and system conditions immediately readable.<\/p>\n\n\n\n

In one invalidation project, the claim described communication between a UE and a base station through a relay node with specific CSI-RS transmission conditions.<\/p>\n\n\n\n

An analyst would normally need close to an hour to fully understand the claim from the text alone. The tool reduced that work to minutes. The diagram surfaced the system architecture, the relay\u2019s role, and the exact conditions governing CSI-RS transmission. Those details shaped the later search strategy.<\/p>\n\n\n\n

The tool also detects issues that text analysis alone may miss.<\/p>\n\n\n\n

\"image\"<\/figure>\n\n\n\n
\"image\"<\/figure>\n\n\n\n

In another project, the tool flagged a contradiction in the phrase \u201cwherein the non-essential minimum SI is a minimum SI.\u201d<\/p>\n\n\n\n

The term \u201cminimum SI\u201d refers to information required for initial cell selection and camping. \u201cNon-essential SI\u201d is not required during that process. The AI agent identified this as a specification failure to clearly define “minimum SI” in context, stronger than a drafting irregularity, and sufficient to open an indefiniteness position that was folded into the combined technical and legal strategy.<\/p>\n\n\n\n

This foundational step produces the precision of claim understanding on which every subsequent search decision depends.<\/p>\n\n\n\t\t

\n\t\t\t
\n\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

Get the Report in your Inbox<\/h4>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
\n\t\t\t\t\t\n\t\t\t\t\t\t\n\t\t\t\t\t\t\t\t\tDownload PDF<\/span>\n\t\t\t\t\t<\/span>\n\t\t\t\t\t<\/a>\n\t\t\t\t<\/div>\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t
\r\n
\r\n