DETAILED ACTION
-Claims 1, 3, 8-10, 13-15 and 17 are amended.
-Objections to the claims are withdrawn based on the claim amendments.
-Rejections under 112(b) are withdrawn based on the claim amendments.
-Claims 1-20 are pending.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant’s Remarks filed on 2/9/2026 have been fully considered.
The arguments regarding the claim objections and 112(b) rejections were found to be persuasive in view of the claim amendments.
The arguments regarding the 103 rejections were not found to be persuasive because Koch [0036] clearly teaches the identification of N most similar sentences based on the similarity scores within the distance matrices, therefore Koch implicitly teaches that such identification of N is based on the distance matrices.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Ayyadurai et al (US Patent No.12,198,030) in view of Koch et al (US Pub. No. 2024/0126991).
Re Claim 1. Ayyadurai discloses a method for matching security recommendation tasks with a regulatory compliance standard, said method comprising: receiving as input at a first Machine Learning (ML) model a regulatory compliance standard (i.e. vector constraints 102 operate as an input into the validation engine 104. The vector constraints 102 can encompass guidelines and/or regulations such as regulatory standards, organizational policies, AI application-specific vector constraints, and industry best practices relevant to the AI application's 106 domain……..using the guidelines input into the validation engine for determining AI compliance, in accordance with some implementations of the present technology. Environment 200 includes guidelines 202 (e.g., jurisdictional regulations 204, organization regulation 206, AI application-specific regulations 208), vector store 210, and validation engine 212. Guidelines 202 can be any of the vector constraints 102……….. The validation engine 212 evaluates the AI application's compliance with the retrieved guidelines 202, (e.g., using semantic search, pattern recognition, and machine learning techniques). For example, the validation engine 212 compares the vector representations of the different explanations and outcomes by calculating the cosine of the angle between the two vectors indicating the vectors' directional similarity. Similarly, for comparing explanations, the validation engine 212 can measure the intersection over the union of the sets of words in the expected and case-specific explanations) [Ayyadurai, (0036, 0044, 0053)]; receiving as input at the first ML model security recommendation tasks (i.e. processes the command set and generates an outcome 310 and explanation 312 on how the outcome 310 was determined based on the AI application's 308 internal algorithms and decision-making processes. The outcome 310 and explanation 312 are evaluated by the assessment module 314, which compares the outcome 310 and explanation 312 against the expected outcomes and explanations specified in the test case 304 derived from the relevant guidelines 302. Methods of evaluating the AI application 308's compliance with the relevant guidelines 302 are discussed in further detail with references to FIGS. 3-11. Any discrepancies or deviations between the observed and expected behavior are flagged as potential compliance issues, warranting further investigation or corrective action) [Ayyadurai, (0060);
Ayyadurai does not explicitly disclose whereas Ayyadurai in view of Koch does: determining by the first ML model distance matrices between the security recommendation tasks and the regulatory compliance standard, wherein the distance matrices define alignments between the security recommendation tasks and the regulatory compliance standard and specify distances therebetween (i.e. the automated interaction processing system can apply a deep leaning model (e.g., BERT model) to generate a first embedding output for portions (e.g., phrases, sentences)……………. the similarity determination component 601 processes the first embedding outputs 606 and the second embedding outputs 608 and outputs similarity scores 610 comprising an N×M dimensional matrix. The similarity scores 610 may comprise values describing a measure of similarity between phrases from the first phrase list 602 and the second phrase list 604. The automated interaction processing system can determine whether each score generated by the similarity determination component meets or exceeds a confidence threshold, such as by meeting or exceeding a predetermined value) [Koch, para.0041, 0058, Fig.6, Note: similarity matrix implicitly teaches distance matrix]; based on the distance matrices, identifying N security recommendation tasks that are within an alignment threshold from the regulatory compliance standard, wherein N is a predetermined number and the alignment threshold is determined based on the distance matrices and N (i.e. the automated interaction processing system can be configured to identify a predetermined number of sentences that are most similar to a target sentence or phrase ) [Koch, para.0036, Note: distance matrices provide the similarity scores and the identifying of N is based on the similarity scores, therefore implicitly, the identifying of N is based on the distance matrices].
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to modify Ayyadurai with Koch because it provides a system for automatically evaluating the performance of agents and customer service representatives ……… the agent or customer representative may be a chatbot or conversational artificial intelligence [Koch, para.0003,0020] and its automated interaction processing system can be configured to identify a predetermined number of sentences that are most similar to a target sentence or phrase e.g. an embedding output or vector such as a numerical representation of at least a portion of content [Koch, para.0036].
Ayyadurai further teaches: generating a prompt, the prompt including the N security recommendation tasks and the regulatory compliance standard; inputting the prompt to a second ML model; and based on the prompt, identifying by the second ML model a subset of the predetermined number N of the security recommendation tasks that match the regulatory compliance standard (i.e. Each of the models is responsible for detecting and assessing specific types of non-compliant content within AI models. Upon processing the training dataset 902, each model generates validation actions tailored to evaluate the presence or absence of specific types of non-compliant content……. The set of generated validation actions 908 is provided as input to an AI application 910 in the form of a prompt. The AI application 910 processes the validation actions 908 and produces an outcome along with an explanation 912 detailing how the outcome was determined. Subsequently, based on the outcome and explanation 912 provided by the AI application 910, the system can generate recommendations 914 for corrective actions. The recommendations are derived from the analysis of the validation action outcomes and aim to address any identified issues or deficiencies. For example, if certain validation actions fail to meet the desired criteria due to specific attribute values or patterns, the recommendations can suggest adjustments to those attributes or modifications to the underlying processes………. if certain attributes exhibit unexpected associations or distributions, the system can retrain the tested AI model with revised weighting schemes to better align with the desired vector constraints. In a toxicity model, such as the ML model discussed in FIG. 6, the corrective actions can include implementing post-processing techniques in the tested AI model to filter out responses that violate the vector constraints (e.g., filtering out responses that include the identified vector representations of the alphanumeric characters). Similarly, in an IP rights violation model, such as the ML model discussed in FIG. 7, the corrective actions can include implementing post-processing techniques in the tested AI model to filter out responses that violate the IP rights) [Ayyadurai, (0129-0131), Note: filtering out responses teaches determining a subset of tasks that match].
Re Claim 8. This claim is similar to claim 1 and is therefore rejected in a similar manner.
Ayyadurai, (00036), as cited above, also teaches standards comprising security tasks.
Re Claim 15. This claim is similar to claim 1 and it is therefore rejected in a similar manner. Ayyadurai, (0129-0131), as cited above, also teaches identifying a subset or a second security recommendation task that does not match.
Re Claims 2, 9 and 16. Ayyadurai in view of Koch discloses the features of claims 1, 8 and 15, wherein the first ML model is a sentence embedding model (i.e. The vector representations can encode the alphanumeric characters into numerical format, allowing the system to process and analyze the vector representations using mathematical operations. To create vector representations of alphanumeric characters, the system can map each alphanumeric character to a dense vector space where semantic similarities between characters are preserved. By leveraging the context of neighboring characters, embeddings capture nuanced relationships between alphanumeric characters. For example, the alphanumeric characters are encoded into vectors using word embeddings. Word embeddings, such as Word2Vec or GloVe, learn vector representations of words based on the word's contextual usage in a large corpus of text data. Each word is represented by a vector in a high-dimensional space, where similar words have similar vector representations) [Ayyadurai, (0090)].
Re Claims 3, 10 and 17. Ayyadurai in view of Koch discloses the features of claims 2, 9 and 16, Ayyadurai in view of Koch discloses further discloses: wherein the sentence embedding model generates embedding vectors for the security recommendation tasks and the regulatory compliance standard (i.e. The validation engine 212 evaluates the AI application's compliance with the retrieved guidelines 202, (e.g., using semantic search, pattern recognition, and machine learning techniques). For example, the validation engine 212 compares the vector representations of the different explanations and outcomes by calculating the cosine of the angle between the two vectors indicating the vectors' directional similarity. Similarly, for comparing explanations, the validation engine 212 can measure the intersection over the union of the sets of words in the expected and case-specific explanations) [Ayyadurai, (0053)],
Koch further discloses: and generates the distance matrices based on the embedding vectors [Koch, as in claim 1].
The motivation to modify with Koch is the same as in claim 1.
Re Claims 4, 11 and 18. Ayyadurai in view of Koch discloses the features of claims 1, 8 and 15, wherein the second ML model is a Large Language Model (LLM) (i.e. In the present disclosure, the term “language model” may be used as shorthand for an ML-based language model (i.e., a language model that is implemented using a neural network or other ML architecture), unless stated otherwise. For example, unless stated otherwise, the “language model” encompasses LLMs) [Ayyadurai, (0158)].
Re Claims 5 and 12. Ayyadurai in view of Koch discloses the features of claims 1 and 8, wherein the predetermined number N is determined by a user of a computing system performing the method (i.e. vector constraints 102 are obtained by manual input by users. For example, users input relevant regulations and policies (e.g. vector constraints) directly into the validation engine through a user interface communicatively connected to the validation engine) [Ayyadurai, (0039)].
Re Claims 6, 13 and 19. Ayyadurai in view of Koch discloses the features of claims 1, 8 and 15, further comprising: normalizing the regulatory compliance standard prior to inputting the regulatory compliance standard into the first ML model (i.e. In some implementations, the guidelines 202 are preprocessed to remove any irrelevant information, standardize the format, and/or organize the guidelines 202 into a structured database schema. Once the guidelines 202 are prepared, the guidelines 202 can be stored in a vector store 210) [Ayyadurai, (0046)].
Re Claim 7. Ayyadurai in view of Koch discloses the features of claims 1, 8 and 15, wherein the subset of the predetermined number N of the security recommendation tasks that match the regulatory compliance standard are used to determine a regulatory compliance score (i.e. The output can include an overall score in accordance with the weighted vector representations of the alphanumeric characters. The system can use a threshold mechanism to determine whether the overall score indicates the presence or absence of certain characteristics or patterns (e.g., toxicity) within the response. For example, if the score exceeds a predefined threshold, the system can classify the response accordingly. There can be multiple thresholds corresponding to different compliance indicators. For example, there can be a threshold for “compliant,” “partially compliant,” and “non-compliant.”) [Ayyudarai, (0102)].
Re Claims 14 and 20. Ayyadurai in view of Koch discloses the features of claims 8 and 15, wherein a title of the security recommendation tasks is input into the first ML model (i.e. Machine learning techniques can be applied to identify patterns or clusters within the guidelines and automatically categorize the guidelines 1002 into relevant scenarios 1004 based on similarity or relevance. Natural Language Processing (NLP) techniques can be used to identify the scenarios 1004 from the guidelines 1002. The system can use named entity recognition (NER), in some implementations, to identify specific terms, phrases, or clauses within the guidelines 1002 that pertain to different scenarios 1004. For example, NER can be used to identify mentions of “data privacy,” “fairness,” “transparency,” “accountability,” or other terms of interest) [Ayyadurai, (0138)].
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to NOURA ZOUBAIR whose telephone number is (571)270-7285. The examiner can normally be reached Monday - Friday.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kambiz Zand can be reached at 571-272-3811. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/NOURA ZOUBAIR/Primary Examiner, Art Unit 2434
Prosecution Insights