SEP Team of LexField Law Offices[1]
During the 2026 IP5 Heads of Office Meeting, the IP5 Offices exchanged views on leveraging artificial intelligence technologies to build an advanced and reliable intellectual property system. As stated in the WIPO Strategy on Standard Essential Patents 2024–2026, assessing the demand for and the feasibility of deploying a voluntary essentiality check service using the pooled resources of IP Offices and, eventually, an AI-assisted element has been identified as a medium-term goal. In light of these developments, for IP practitioners, using artificial intelligence (AI) to assist in patent analysis is no longer merely a tool for improving efficiency, but a foundational skill that must be mastered. This article discusses the challenges of standard-essential patent (SEP) analysis and the advantages and limitations of AI-powered tools, and proposes which tasks should be performed by AI-powered tools and which by human experts.
I. Introduction to Standard Essential Patent Analysis
A standard essential patent (SEP) is a patent that protects an invention essential to the implementation of a particular technology standard. Standards include international standards, national standards, and industry standards, among others.
In general terms, the standard-essentiality analysis of a patent involves determining whether the technology protected by the patent is "essential" when a product implements a particular standard. Specifically, each technical feature of at least one claim of the patent—not merely the technical concept or design idea recorded in the patent document—must map to a technical solution in a standard implemented by the product. "Mapping to a technical solution in the standard" is not merely a high degree of textual similarity, but rather based on the understanding of the claims of a person having ordinary skill in the art after reading the patent, the technical solution in the standard falls within the scope of protection of the claim.
Determining whether a patent is an SEP can be a difficult process that requires multiple time-consuming steps, including at least the following:
1. Divide the claim into technical features and compare each technical feature with the relevant technical standard specification.
2. Determine whether the terms in the technical features of the claim are identical or equivalent to the terms in the technical standard specification. This requires consideration of the patent specification, patent prosecution history, priority documents, family history, past infringement and invalidation litigation histories, as well as background and prior art, and in accordance with relevant laws and regulations regarding claim construction[2]. A rigorous and definitive yes or no conclusion must be reached, rather than a probabilistic score of how likely the features are identical or equivalent.
3. Confirm that the portions of the standard to which the claim features are mapped (if they are not contiguous) coherently fit together to be a technical solution describing some common aspect or operation of the system, rather than being merely a set of disjointed clauses.
4. Confirm that the technical solution of the standard to which the claim is mapped, as described in step 3, is actually required for compliance with the standard. When a patent maps to an optional feature of a standard, showing that the optional feature is practiced requires additional effort by providing product use evidence, such as technical manuals or test evidence.
As can be seen, patent standard-essentiality analysis is a highly specialized skill that requires simultaneous mastery of law and technology, and is a challenging, time-consuming task that requires a skilled practitioner to spend multiple hours, at a minimum, and potentially a day or more. In the technical negotiations of SEP licensing transactions, experts from both parties communicate and debate over whether a particular patent is a true "Standard Essential Patent (TRUE SEP)," typically requiring at least three rounds of technical meetings—the patent holder presents the claim chart, the standard implementer raises rebuttal opinions, and the patent holder rebuts the standard implementer's rebuttal opinions. In patent infringement litigation, judges need to go through complex and lengthy procedures to provide legal interpretations of disputed terms in the patent claims, clarify the scope of protection of the patent, and then compare the technical standard specification with the technical features of the patent claims one by one to confirm whether the technical solution in the standard falls within the scope of patent protection.
II. Challenges in Standard Essential Patent Analysis
In the technical negotiation and commercial negotiation steps of the good-faith negotiation process for SEP licensing transactions, the patent holder should provide the standard implementer with a list of standard essential patents, a reasonable number of claim charts, the calculation method and basis for the royalty rate, a reasonable feedback period, and other particulars. The standard implementer should accept the licensing terms within a reasonable period; if not, it should provide a counteroffer that complies with the fair, reasonable, and non-discriminatory (FRAND) principles within a reasonable period[3]. In the process of both parties endeavoring to meet the aforementioned good-faith negotiation requirements, the challenges they face in standard essential patent analysis include:
1. SEP asset verification and technical assessment: the standard implementer needs to analyze which and how many of the patents in the list provided by the rights holder are true SEPs with stable validity, to serve as the basis for a counteroffer. A patent list of SEP licensing transactions between large enterprises typically includes several hundreds to over a thousand patents.
2. Top-down approach for calculating royalty rate: calculating the industry aggregate royalty rate involves estimating the total number of TRUE SEPs in the industry and the proportion of TRUE SEPs held by each patent holder in the industry. These data are neither publicly available nor transparent.
3. Analysis of Standard-Setting Organization (SSO) SEP declaration data:
In the wireless communications field, the problem of over-declaration of standard essential patents is severe. LexisNexis 5G report says the number of 5G patent families declared to ETSI currently exceeds 66,000[4]; however, the number of patents related to standards far exceeds the number of standard essential patents. Current SEP analysis reports on the market vary greatly in their results due to different sampling methods.In the audio and video codec fields, the problem of missing standard essential patent declaration information is relatively prominent. Although the intellectual property policies of international SSOs such as ITU/ISO/IEC encourage members to disclose early and to the best of their ability, they do not mandate members to list their SEPs one by one. Therefore, it is common to find declaration forms in the ITU online database that are "not accompanied by specific patent numbers."[5]
In SEP infringement and invalidation litigation, the dilemma faced by patent holders is that arguments and amendments in invalidation proceedings must maintain the validity of the patent while preserving its standard essentiality. Challenges faced by standard implementers include how to raise a non-infringement defense in infringement litigation, and whether to challenge the validity of the patent through invalidation proceedings and how to select patents for filing invalidation requests.
Ⅲ. What Tasks in SEP Analysis Can Be Handled by AI
Currently, popular AI-powered patent tools in China include PatSnap, incoPat, BaYueGua, QiZhiDao, etc. Foreign AI-powered patent tools include LexisNexis IPlytics, Patlytics, IP Mind, PatSeer, Orbit Intelligence, Derwent Innovation, Patently, Cypris, Perplexity Patents, DeepIP, as well as free tools such as Google NotebookLM, Lens.org, and ChatGPT. The majority of these tools can provide AI-powered intelligent patent interpretation, claim feature analysis, and comparison of patent and product/standard technical solutions.
Dr. Tim Pohlmann of IPlytics has written about AI-based SEP prediction models that score patents as to their likelihood of being standard-essential, and concluded that while AI-based SEP determination may not replace the work of experts, it supports valuating and determining the essentiality of SEPs for various use cases such as patent portfolio management, patent licensing and transactions, and economists' valuation of patent portfolios in the course of a top-down analysis.[6]
Patlytics purports that its SEP check module can automatically compare a single patent with a target standard protocol and generate a claim chart within 15 minutes, and provide an essentiality score. R&D departments can determine whether their innovations relate to standardized functionality. Licensing teams can identify patents that may hold strategic value or require further evaluation in negotiation contexts. Portfolio managers can benchmark essentiality evidence against competitor portfolios.[7]
As mentioned earlier, conducting standard essential patent analysis is a costly and time-consuming task. This is particularly true in the wireless communications field, where analyzing the hundreds to thousands of patents in an SEP licensing transaction patent list, or the tens of thousands of declared patents in the ETSI online database by experts, is an unrealistic task. Since AI-powered tools can quickly provide probabilistic scores based on the semantic similarity between standard specification texts and patent texts, they can leverage the efficiency in the preliminary screening of TRUE SEPs.
IV. Limitations of Current AI-powered Tools in SEP Analysis
A 2022 research paper by Professor Katie Atkinson and Professor Danushka Bollegala of the Department of Computer Science at the University of Liverpool, UK[8], surveyed the existing professional AI-powered tools in the domain of law and commercial AI-powered tools available for patent essentiality reviews. Their conclusion was that, due to the complexity of standard essentiality determinations, current techniques and tools cannot replicate or replace human expert review. The paper identified the following three reasons:
1. Existing industry-leading AI solutions for the automatic detection of SEPs measure the semantic similarity between a given standard specification and a patent. However, similarity and essentiality are not equivalent concepts. A patent might be essential to a standard but might not necessarily have a high similarity in terms of textual overlap. On the other hand, between two patents that are highly similar to a given standard, one could be essential while the other might not.
2. The same words and phrases, with only a different relative ordering of words within the text, may have significant legal meaning distinctions, but AI tools still have shortcomings in this analytical capability.
3. The exact definition of what is an essential patent for a given standard remains a subjective one. Two different patent attorneys often disagree on the same set of patents being essential, as evidenced by litigation. In the machine learning community, this is known as inter-annotator agreement, which refers to whether the judgment results of different experts are consistent and reliable when performing SEP essentiality annotations. There are established technical measures for measuring the inter-annotator agreement for subjective annotating tasks. However, to the best knowledge of the authors of this article, such inter-annotator agreements are not deployed in AI-powered tools for SEP detection.
On May 6 of this year, Harvey, an AI company focused on the legal field, open-sourced its internal benchmark for evaluating AI legal agents—the Legal Agent Benchmark (LAB)—on a public website. Another independent AI benchmarking institution, Artificial Analysis, evaluated Harvey's LAB and announced that it had tested 28 AI tools, including Claude Fable 5, on 120 real-world legal tasks across 24 practice areas. The highest overall pass rate for these AI tools completing all 120 real-world legal tasks was only 14.2%. That is, the probability that these AI tools would correctly complete an entire legal task without missing a single standard was only one in seven.[9]
From this, it can be inferred that current AI-powered tools are not yet competent in legal tasks such as patent claim construction and determining the scope of claim protection. Therefore, it is better to have experts performing this work first, clarifying the technical solution protected by the patent claims, and then re-describing the technical solution protected by the patent claims in ordinary technical language (rather than the legal language used in patents) before inputting it into AI for essentiality determination.
V. Conclusion
AI-powered tools may leverage efficiency significantly in tasks such as the rapid preliminary assessment of large-scale standard essential patent portfolios. However, in legal tasks such as patent claim construction, the analytical results by AI are still not reliable. It is recommended that this portion of the work be assigned to experts.
[1] Author：Wen Zhou, M.S. in Engineering, Beijing University of Posts and Telecommunications, Of Counsel, LexField Law Offices.
[2] For provisions on claim construction, see the Chinese Patent Law, Implementation Regulations of the Chinese Patent Law, Patent Examination Guidelines, Interpretation and Interpretation (II) of the Supreme People's Court on the Application of Laws in the Trial of Disputes Regarding the Infringement of Patent Rights, Provisions (I) of the Supreme People's Court in the Trial of Cases Concerning Patent Grant and Confirmation, and Guidelines for Patent Infringement Determination (2017), Beijing High People’s Court, among other relevant laws and regulations.
[3] See Chapter II Article 8 of the Anti-Monopoly Guidelines for Standard Essential Patents issued by the State Administration for Market Regulation on November 4, 2024.
[4] See LexisNexis: Who Is Leading the 5G Patent Race? 5G Report - Janurary 2026 https://www.lexisnexisip.cn/5g-report-2026/ ，last visit date 2026-7-16.
[5] See LexField Law Offices: Analysis of the Background and Characteristics of Standard Essential Patent Licensing in Video Codec Field. https://www.lexology.com/library/detail.aspx?g=e85b81bb-67e7-4436-ad27-a7f89809e7e0 , publication date 2026-01-06，last visit date 2026-7-16
[6] See Tim Pohlmann: Using AI to Valuate and Determine Essentiality for SEPs https://ipwatchdog.com/2021/06/18/using-ai-valuate-determine-essentiality-seps/# ，publication date：2021-06-18，last visit date：2026-7-16.
[7] See Patlytics: How AI Software Speeds Up SEP Essentiality Review and Analysis https://www.patlytics.ai/blog/how-ai-software-speeds-up-sep-essentiality-review-and-analysis，last visit date：2026-7-16.
[8] See Katie Atkinson & Danushka Bollegala：AI for Patent Essentiality Review https://livrepository.liverpool.ac.uk/3168569/1/AI_for_Patent_Essentiality_Review.pdf ，publication date：2022-11，last visit date：2026-7-16.
[9] See Artificial Analysis: Announcing Harvey LAB-AA: evaluating AI agents on real-world legal work https://artificialanalysis.ai/articles/harvey-lab-aa ，publication date：2026-07-07，last visit date：2026-7-16.
