DETAILED ACTION
This Final Office Action is in response to amendment filed on 08/04/2026. Claims 6 and 16 have been amended. Claims 1-20 remain pending in the application.
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 Objections
Applicant’s amendments to the claims filed on 08/04/2026 have been reconsidered. Applicant’s amendments to the claims have been accepted. Objections to the claim, previously set forth in the office action mailed on 05/07/2026, are withdrawn.
Drawings
The drawings filed on 01/14/2025 are accepted.
Response to Arguments
Applicant's arguments filed 08/04/2026 have been fully considered but they are not persuasive.
Applicant stated that (a) Baker does not disclose "processing, by a first artificial intelligence (AI) computer model, features of the detected code change as input to generate a classification of the code change as being either functional or non-functional," because Baker's generative AI system is used only to generate code-change explanation responses; and (b) Monea's classification of source code as malicious or non-malicious is not the same classification as "functional or non-functional" in pages 9-10.
Examiner respectfully disagrees. (a) Regarding Baker: Applicant's argument is directed to Baker in isolation. It is settled that nonobviousness cannot be established by attacking references individually where, as here, the rejection is based on a combination of references. The rejection relies on Monea, not Baker, for the classification step and its underlying feature-based determination. (Monea [0027], [0030].) Baker is relied upon solely for the proposition that implementing such a classification function using an "artificial intelligence (AI) computer model," as opposed to some other implementation, was well known and predictable in the art. That Baker's own generative AI system happens to be used for a different end purpose (generating explanation responses) does not undermine the stated rationale for the combination, which is that it would have been obvious to implement Monea's code-feature classification architecture as an AI computer model of the type Baker discloses, in order to "strengthen identifying patterns in data." (Baker [0040].) This portion of Applicant's argument is accordingly not persuasive because it addresses Baker in isolation rather than the combined teachings of Monea and Baker.
b) Regarding the "functional or non-functional", Examiner respectfully disagrees. Given the broadest reasonable interpretation, a system that extracts features from code and uses those features to classify the code's behavior falls within the scope of the claimed classification step. Classifying a code change as "functional" or "non-functional" is, under this interpretation, simply a different labeling of the same type of binary output Monea already discloses as "malicious" or "non-malicious". The claimed binary distinction is accordingly an obvious variation of the reference's binary determination. This reading is consistent with Applicant's own specification, which frames the invention around this same benign/malicious distinction:
([0018] “Another drawback is that a large volume of alerts may be generated when resources are changed and those changes are made through benign processes. That is, the SRI based monitoring services are not able to distinguish between changes that are benign and those that are malicious. Thus, there is a need for an improved computing tool and improved computing tool operations/functionality that addresses these drawbacks in existing technology. The illustrative embodiments provide computing mechanisms that address these drawbacks and provide CMS level capabilities to provide SRI generation that cannot be easily fooled by attackers and which distinguishes between benign and malicious changes so as to minimize the generation of alerts”).
For these reasons, the rejection of claim 1 is maintained. Claims 11 and 20 recite substantially the same limitations as claim 1 and are rejected for the same reasons; Applicant's arguments as to claims 11 and 20 are not persuasive for the reasons given above.
Applicant additionally stated, specifically as to claim 2, that Monea's disclosure of associating a hash of a file with a natural-language description and a malicious/non-malicious label, and of retraining/fine-tuning a classification model, does not disclose "updating a hash value of the code of the resource to be a new hash value corresponding to the changed code of the resource" in response to a functional and benign classification in page 10-11.
Examiner respectfully disagrees. Monea discloses a results database that associates a hash of a file with an NL description and a label indicating whether the file is malicious or non-malicious, and that this association is generated and stored by the CSM agent for each processed file. (Monea [0035].) Once the CSM agent determines that a given file's current label does not correspond to a malicious classification, the hash currently associated with, and stored for, that file in the results database is the hash of the current (changed) code — i.e., the system of record's hash for the file is thereby updated to correspond to the changed, non-flagged code. (Monea [0035], [0039].) This is the functional equivalent of the claimed hash update performed in response to a functional-and-benign determination. Applicant's characterization of this disclosure as merely describing static training-data association does not address that the CSM agent's association-and-storage operation necessarily updates the recorded hash each time a new, non-flagged version of a file is processed.
Applicant additionally stated, that Grover's disclosure of comparing a runtime thread pattern to a learned thread pattern to identify variances (Grover [0044]), and its unsupervised-learning embodiment for identifying patterns "different from the others" (Grover [0047]), is directed to anomalous runtime execution behavior (e.g., memory usage of executing threads) and does not teach or suggest determining, based on extracted features of a natural-language narrative, whether a code change deviates from historical patterns of code changes to resources, as recited in claim 6 in pages 12-13.
Examiner respectfully disagrees. Although Grover's cited passages, read in isolation, describe comparison of runtime thread-execution patterns rather than analysis of a natural-language code-change narrative. The rejection, however, does not rely on Grover for a disclosure of narrative-based analysis in isolation. The combination relies on Monea and Baker for the third AI computer model's narrative-based feature extraction (Monea [0013], [0043]) and relies on Grover for the more general technique that a deviation of observed data from a learned or historical baseline pattern — assessed through feature comparison, including via unsupervised techniques that identify patterns "different from the others" without requiring a known baseline (Grover [0047]) is indicative of malicious activity. Applying that known deviation-from-baseline analytical technique to the historical code-change-pattern comparison already performed by the combined system on the third AI model's narrative-derived features is a predictable application of a known analytical technique to an analogous problem (detecting anomalous or malicious behavior from a stream of extracted feature data), yielding predictable results. Applicant's arguments address only whether Grover's specific runtime-thread embodiment discloses the claimed narrative-based comparison; and does not address the Examiner's rationale for combining Grover's deviation-detection technique, set forth above, with the narrative-based feature extraction already taught by the Monea/Baker combination.
Because Applicant's arguments regarding claims 1 and 11 and 20 are not persuasive for the reasons discussed above, and Applicant's Grover-specific arguments do not address the combination as actually articulated, the rejection of claims 6, 7, 16, and 17 is maintained.
Applicant's arguments as to claims 8-10, 18, and 19 rest entirely on the dependency of these claims from claims 1 and 11. Because the rejections of claims 1 and 11 are maintained for the reasons set forth above, the rejections of claims 8, 9, 10, 18, and 19 are likewise maintained.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
Claim(s) 1,2,3,4,5,11,12,13,14,15, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Monea (US 20250217473A1) in view of Baker (US20250272091A1).
Regarding claim 1, Monea discloses a method (Abstract: “The present disclosure relates generally to cybersecurity, and more particularly, to systems and methods of using generative artificial intelligence (AI), such as large language models (LLMs), to generate natural language descriptions of code (e.g., source code, executable code) for enhanced threat analysis and malware detection”) comprising:
detecting a code change in code of a resource; processing, by a [first artificial intelligence (AI) computer model], features of the detected code change as input to generate a classification of the code change as being either functional or non-functional (“[0027] The first set of CP training data may include event data and files, where each file includes computer code of at least the first type (e.g., Java). The first set of CP training data may also include ground truth data (e.g., labels) indicating whether each file is malicious or non-malicious. A file is deemed to be malicious if it includes or references code that is designed, when executed by a computer, to waste computer resources (e.g., memory, hard drive, processing, networking), harm the resources of a computer, and/or cause one or more computers to perform deceptive operations”); [0030] “For example, the NL description may indicate that the classification platform 114 assigned a maliciousness score of 80% to the file and that the CSM system 106 (e.g., the CSM agent 108 or the classification platform 114) determined that the 80% falls within the range corresponding to the high threshold band. In some embodiments, a high threshold means that the given sample has the highest probability of being correctly classified as malicious”; a system that extract features from code and uses those features to classify the code’s behavior falls within that concept, note that classification of “functional/non-functional” is just changing the output categories as “malicious” or “non-malicious”, meaning the claimed binary distinction is an obvious variation of the reference’s binary determination, see Applicant’s specs paragraph [0018] for defining functional/nonfunctional; Note that the context of the designations in the reference does not carry an ordinal meaning. The code platform may include on or more different types of computer code, a NL description dataset. The numerical designations are reference numerals used to distinguish between instances of LLM component – See reference paragraph [0028];
in response to the code change being classified as functional: generating, by a second Al computer model, a natural language narrative description of the code change (([0028] “The code explainer platform 116 includes and/or executes LLMs 117 (e.g., LLM 117a, LLM 117b, and LLM 117c) for generating natural language explanations for computer code (e.g., source code, executable code, and/or object code). Specifically, the code explainer platform 116 includes an LLM 117a that is trained to generate, based on a file including computer code of a first type (e.g., Java), a NL description dataset that provides a NL description of the file”); [0061] “The method 300 includes the block 302 of determining that a file comprises source code for causing malicious activity. The method 300 includes the block 304 of generating, by a processing device and using one or more large language models (LLMs), a natural language (NL) description of the source code responsive to determining that the file comprises the source code for causing the malicious activity”), note that the second AI computer model is interpreted as the reference’s code explainer platform 116, executing LLMs 117, to perform various operations to generate a code change explanation response and provide the response to a requestor of a code change explanation request. The code change explanation response comprises a natural language narrative description of the code change, generated by one or more language models, LLMs – see Reference paragraph [0028]; and
processing, by a third Al computer model, the natural language narrative to extract features based on a computer natural language processing (NLP) of the natural language narrative to determine a threat level of the code change (classification platform 118, based on the provided NL as disclosed in [0043], determines different types/pattern/feature of malicious activities, e.g. [0013] " where each maliciousness score indicates whether source codes are associated with one or more types (e.g., specific type of calls) of malicious activity. For example, the maliciousness score indicates that the source code is associated with the one or more types of malicious activity or the source code fails to meet the criteria for being associated with the one or more types of malicious activity”; [0043] “The CSM agent 106 may be configured to provide the NL descriptions of the source code to a classification model (e.g., enhanced classification platform 118) trained to generate a first set of maliciousness scores each indicating whether source codes are associated with one or more types of malicious activity. The CSM agent 106 may be configured to generate, using the classification model (e.g., enhanced classification platform 118), a maliciousness score for the source code indicating that the source code is associated with the one or more types of malicious activity”; Note that the third AI computer model is interpreted as the reference’s Classification Platform 118 to generate a threat level, second set of maliciousness scores, each indicating whether the source codes is associated with malicious activity, where the score/ threat level is determined based on the provided NL and processed NL using the classification platform 118; and
registering or denying registration of the code change in a content management system (CMS) based on the threat level of the code change ([0032] “The enhanced classification platform 118 sends a message to the CSM agent 108, where the message includes the file (or file identifier) and the maliciousness score (shown in FIG. 1 as, “maliciousness score 2”) that is assigned to the file. The CSM agent 108 stores the maliciousness score and linked file (or file identifier) in the results database 112. The classification platform 114 also stores the file in the source code file database 111.; CSM stores or registers the code change in the database 112 or CMS; [0039] “FIG. 2A is a block diagram depicting an example of the cybersecurity management (CSM) system of the environment in FIG. 1 , according to some embodiments. While various devices, interfaces, and logic with particular functionality are shown, it should be understood that the CSM system 106 includes any number of devices and/or components, interfaces, and logic for facilitating the functions described herein”; ) registering or denying registration of the code change in a content management system (CMS) based on the threat level, maliciousness score, of the code change ; content management system is associated with a group of computing systems that are under management.
However, Monea does not explicitly disclose the first artificial intelligence (AI) computer model. Baker further discloses the first artificial intelligence (AI) computer model ([0025] “In FIG. 1 , generative AI system 104 comprises at least one API 114 and language model 120. In other implementations, language model 120 is not included in generative AI system 104 and generative AI system 104 is in communication with language model 120 via one or more APIs”), note that the first AI computer model is interpreted as a generative AI system in reference Baker.
Monea and Baker are analogous in systems that uses AI computing systems. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Baker to implement a first artificial computer model. Doing so would strengthen identifying patterns in data and determine appropriate course of action (Baker [0040]” Typically, larger numbers of parameters result in a more complex language model 120 that has a strong understanding of the structure and meaning of data, which enables language model 120 to efficaciously identify intricate patterns in the data” )
Regarding the computer program product claim 11 and computer system claim 20, the claim recites similar limitations as the computer method claim 1, therefore, rejected based on the same rationale as claim 1.
Regarding claim 2, the combination of Moneo and Baker teaches the first, second, and third AI computers. Monea further discloses in response to the code change being classified as functional, and in response to determining that the threat level of the code change is benign, updating a hash value of the code of the resource to be a new hash value corresponding to the changed code of the resource ( [0035] “The CSM agent 108 associates several datasets together and stores the associated data in the results database 112. For example, as shown in FIG. 1, the results database 112 includes a first dataset that associates a hash (e.g., hash 1) of file 1 with a NL description 1 of file 1 and a label 1 indicating whether file 1 is malicious or non-malicious; a second dataset that associates a hash (e.g., hash 2) of file 2 with a NL description 2 of file 2 and a label 2 indicating whether file 2 is malicious or non-malicious; a third dataset that associates a hash (e.g., hash 3) of file 3 with a NL description 3 of file 3 and a label 3 indicating whether file 3 is malicious or non-malicious; up to an N-th dataset that associates a hash (e.g., hash N) of file N with a NL description N of file N and a label N indicating whether file N is malicious or non-malicious”; [0050] "The CSM agent 106 may be configured to retrain (e.g., update, refine, fine-tune), using the plurality of NL descriptions, the classification model to generate a second set of maliciousness scores each indicating whether the source codes are associated with the one or more types of malicious activity. The CSM agent 106 may be configured to retrain, using the plurality of NL descriptions, the one or more LLMs to generate a set of NL descriptions of source codes. In some embodiments, the CSM agent 106 may be configured to retrain or fine-tune 1 total reaction.", where the update/fine-tuning of the CSM includes the storing of the hash of the results in the DB 112), a system that extract features from code and uses those features to classify the code’s behavior falls within that concept, note that classification of “functional/non-functional” is just changing the output categories as “malicious” or “non-malicious”; thus, updating a hash value of the code corresponding to the changed code.
Regarding the computer program product claim 12, the claim recites similar limitations as the computer method claim 2, therefore, rejected based on the same rationale as claim 2.
Regarding claim 3, the combination of Moneo and Baker teaches the first, second, and third AI computers. Baker further teaches wherein detecting a code change in code of a resource comprises executing a difference command on code in an inventory catalog to extract line-by-line changes in the code ([0028] “In some implementations, the code change explanation request includes one or more identifiers (e.g., a software code file identifier, line number identifier, method name, and/or function name) corresponding to the user request… Examples of information sources include a current and previous versions of a software code file, a line-by-line history of changes, commits included in a repository commit history database associated with the code change, pull requests included in a pull request database associated with the commits, issues recorded in an issue database associated with the pull requests, associated work items in a work item database, design documents stored in a document management system related to a project of which the software code file is included, and/or a collaborative webpage related to the software code project (e.g., a webpage including collaborative content)”; [0065] “Another example information source 210 is a line-to-line history of changes 308 of the software code file 310. For instance, the line-to-line history of changes 308 includes information about a change made to a line of software code, such as a unique commit identifier 314 of a commit 301 that introduced the change, and other metadata (e.g., person who made the change, services or application used to make the changes, timestamps). In examples, generative AI system 104 may obtain the commit identifier 314 to collect information about the commit 301 and other related commits 301. For instance, generative AI system 104 may execute a search using repository commit history log 312 based on the commit identifier 314 to collect a natural language commit message 316 describing the change and other commit details 318. The repository commit history log 312 provides an overview of the software code repository's commit history. For instance, a chronological list is provided of commits 301 made to the entire software code repository 110 or to a specific branch including information about each commit 301, such as the commit identifier 314, commit messages 316, commit details 318, and/or other information about the changes made in each commit 301”, Note that the system analyzes changes when code is “checked in” to identify exactly what has changed, down to language-specifics such as method bodies, class interfaces, and method signatures”; thus, this aligns with extracting specific, detailed changes from the code; Note that “source code repository” is interpreted as “inventory catalog”.
Monea and Baker are analogous in systems that uses AI computing systems. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Baker to implement an explanation about a function change. Doing so would strengthen identifying patterns in data and determine appropriate course of action (Baker [0040]” Typically, larger numbers of parameters result in a more complex language model 120 that has a strong understanding of the structure and meaning of data, which enables language model 120 to efficaciously identify intricate patterns in the data” ).
Regarding the computer program product claim 13, the claim recites similar limitations as the computer method claim 3, therefore, rejected based on the same rationale as claim 3.
Regarding Claim 4, the combination of Moneo and Baker teaches the first, second, and third AI computers. Monea further discloses wherein the second Al computer model generates the natural language narrative description based on a classification of the code change ([0015] “The CSM system 106 includes and/or executes an CSM agent 108, a classification platform 114, a code explainer platform 116, an enhanced classification platform 118, an event database (DB) 110, a source code database 111, a results database 112. In some embodiments, the event database 110, the source code database 111, and/or the results database 112 may be an elastic search cluster”; [0045] “The CSM agent 106 may be configured to determine that the file includes source code for causing malicious activity without executing executable code associated with the source code. The CSM agent 106 may be configured to generate, using the one or more LLMs, the NL descriptions of the source code, but without executing the executable code associated with the source code”; [0049] “The CSM agent 106 may be configured to maintain, in a database (e.g., NL description database 112), a plurality of NL descriptions associated with a plurality of file identifiers. Each NL description of the plurality of NL descriptions is respectively associated with a respective file identifier of the plurality of file identifiers”), note that the computer model generates the native language description based on the category changed by code change, interpreted as the user request, along with the context information about the function change regarding the code, interpreted as identifiers in a database, e.g. database 112; note that the second AI computer model is interpreted as the reference’s code explainer platform 116, executing LLMs 117, to perform various operations to generate a code change explanation response and provide the response to a requestor of a code change explanation request. The code change explanation response comprises a natural language narrative description of the code change, generated by one or more language models, LLMs – see Reference paragraph [0028].
However, Monea does not explicitly teach the limitation listed below. Baker further discloses context information about a type of function changed by the code change ([0004] “In some implementations, the code change explanation request includes one or more identifiers (e.g., a software code file identifier, line number identifier, method name, and/or function name) corresponding to the user request. For instance, one or more software code identifiers specify or otherwise indicate the software code file and/or lines, method, function, or other portion of the software code file including the change of interest associated with the user request”) Note that context information about a function changed by code change is interpreted as operations to analyze change to functions by the code change.
Monea and Baker are analogous in systems that uses AI computing systems. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Baker to implement an explanation about a function change. Doing so would strengthen identifying patterns in data and determine appropriate course of action (Baker [0040]” Typically, larger numbers of parameters result in a more complex language model 120 that has a strong understanding of the structure and meaning of data, which enables language model 120 to efficaciously identify intricate patterns in the data” ).
Regarding the computer program product claim 14, the claim recites similar limitations as the computer method claim 4, therefore, rejected based on the same rationale as claim 4.
Regarding claim 5, the combination of Moneo and Baker teaches the first, second, and third AI computers. Baker further discloses the method of claim 1, wherein [the second AI computer model] is a transformer-based natural language processing AI computer model ([0039] “In some examples, language model 120 is implemented using a neural network, such as a deep neural network, that utilizes a transformer architecture to process received input”).
Monea and Baker are analogous in systems that uses AI computing systems. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Baker to implement a transformer-based natural language computer model. Doing so would strengthen identifying patterns in data and determine appropriate course of action (Baker [0040]” Typically, larger numbers of parameters result in a more complex language model 120 that has a strong understanding of the structure and meaning of data, which enables language model 120 to efficaciously identify intricate patterns in the data” )
Regarding the computer program product claim 15, the claim recites similar limitations as the computer method claim 5, therefore, rejected based on the same rationale as claim 5.
Claim(s) 6,7, 16, and 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Monea (US 20250217473A1) in view of Baker (US20250272091A1) and Grover (US 20240086522 A1).
Regarding claim 6, the combination of Monea and Baker teaches the first, second, and third AI computers; wherein the third AI computer model determines based on the extracted features from the natural language narrative (classification platform 118, based on the provided NL as disclosed in [0043], determines different types/pattern/feature of malicious activities, e.g. [0013] " where each maliciousness score indicates whether source codes are associated with one or more types (e.g., specific type of calls) of malicious activity. For example, the maliciousness score indicates that the source code is associated with the one or more types of malicious activity or the source code fails to meet the criteria for being associated with the one or more types of malicious activity”; [0043] “The CSM agent 106 may be configured to provide the NL descriptions of the source code to a classification model (e.g., enhanced classification platform 118) trained to generate a first set of maliciousness scores each indicating whether source codes are associated with one or more types of malicious activity. The CSM agent 106 may be configured to generate, using the classification model (e.g., enhanced classification platform 118), a maliciousness score for the source code indicating that the source code is associated with the one or more types of malicious activity”).
However, Monea does not explicitly teach the listed limitation below. Grover further discloses whether the code change deviates from historical patterns of code changes to resources and classifies the code change to be malicious when the code change deviates from historical patterns of code changes (“[0044] The node/link information may also be used when comparing the thread pattern 210A to the learned thread pattern 122A to identify variances. The comparison may use thresholds when identifying a malicious thread pattern based on one or more of these factors. For example, if the memory usage is dramatically different for a thread 201 of the thread pattern 210A versus the learned thread pattern 122A, this may indicate a malicious behavior or some other issue with the thread 201/application 106”; [0047] “In one embodiment, where unsupervised machine learning is used, the process can look for patterns that are different from the others. In this embodiment, it doesn't require comparison to a known good baseline pattern”), Note that the “deviation from patterns” teaches the concept of threat detection through historical analysis, wherein the system compares changes between versions.
Monea, Baker, and Grover are analogous in systems that detects source code changes. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Kim to implement known change patterns and determining a threat level. Doing so would strengthen identifying changes in code and determine appropriate course of action (Grover [0002] “Identification of malicious attacks and malware are an ongoing issue in computer networks/computer systems. As malicious attacks/malware evolve, identification of new types has always been difficult to detect. Typically, it may require multiple malware detection processes and techniques to identify a new type of malicious attack/malware”).
Regarding the computer program product claim 16, the claim recites similar limitations as the computer method claim 6, therefore, rejected based on the same rationale as claim 6.
Regarding claim 7, the combination of Monea and Baker teaches the first, second, and third AI computer model as listed above. Monea further teaches the second AI computer model is trained through a machine learning process, on training data comprising examples of narratives and corresponding ground truth labels specifying a proper classification of the changes described in the examples of narratives ([0015] “The CSM system 106 includes and/or executes an CSM agent 108, a classification platform 114, a code explainer platform 116, an enhanced classification platform 118, an event database (DB) 110, a source code database 111, a results database 112. In some embodiments, the event database 110, the source code database 111, and/or the results database 112 may be an elastic search cluster”; note that the second AI computer model is interpreted as the reference’s code explainer platform 116, executing LLMs 117, to perform various operations to generate a code change explanation response and provide the response to a requestor of a code change explanation request. The code change explanation response comprises a natural language narrative description of the code change, generated by one or more language models, LLMs – see Reference paragraph [0028]; [0034] “The first set of CP training data may include event data and files, where each file includes computer code of at least the first type (e.g., Java). The first set of CP training data may also include ground truth data (e.g., labels) indicating whether each file is malicious or non-malicious”; [0035] “The CSM agent 108 associates several datasets together and stores the associated data in the results database 112. For example, as shown in FIG. 1 , the results database 112 includes a first dataset that associates a hash (e.g., hash 1) of file 1 with a NL description 1 of file 1 and a label 1 indicating whether file 1 is malicious or non-malicious; a second dataset that associates a hash (e.g., hash 2) of file 2 with a NL description 2 of file 2 and a label 2 indicating whether file 2 is malicious or non-malicious; a third dataset that associates a hash (e.g., hash 3) of file 3 with a NL description 3 of file 3 and a label 3 indicating whether file 3 is malicious or non-malicious; up to an N-th dataset that associates a hash (e.g., hash N) of file N with a NL description N of file N and a label N indicating whether file N is malicious or non-malicious”) this model is trained on training data for models that process code narratives; the method consist of messages (narratives) paired with their classifications (truth labels) “malicious” or “non-malicious” to train a machine learning model;
wherein [the first AI computing model] is trained to recognize patterns of input features that correspond to at least one of malign or benign changes to code ((“[0027] The first set of CP training data may include event data and files, where each file includes computer code of at least the first type (e.g., Java). The first set of CP training data may also include ground truth data (e.g., labels) indicating whether each file is malicious or non-malicious. A file is deemed to be malicious if it includes or references code that is designed, when executed by a computer, to waste computer resources (e.g., memory, hard drive, processing, networking), harm the resources of a computer, and/or cause one or more computers to perform deceptive operations)”), note that the first AI computer model is interpreted as a generative AI system in reference Baker.
Monea does not explicitly disclose the first artificial intelligence (AI) computer model. Baker further discloses the first artificial intelligence (AI) computer model ([0025] “In FIG. 1 , generative AI system 104 comprises at least one API 114 and language model 120. In other implementations, language model 120 is not included in generative AI system 104 and generative AI system 104 is in communication with language model 120 via one or more APIs”), note that the first AI computer model is interpreted as a generative AI system in reference Baker.
Monea and Baker are analogous in systems that uses AI computing systems. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Baker to implement a first artificial computer model. Doing so would strengthen identifying patterns in data and determine appropriate course of action (Baker [0040]” Typically, larger numbers of parameters result in a more complex language model 120 that has a strong understanding of the structure and meaning of data, which enables language model 120 to efficaciously identify intricate patterns in the data” )
Regarding the computer program product claim 17, the claim recites similar limitations as the computer method claim 7, therefore, rejected based on the same rationale as claim 7.
Claim(s) 8 and 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Monea (US 20250217473A1) in view of Baker (US20250272091A1) and Kim (US 20230252145A1).
Regarding claim 8, the combination of Monea and Baker teaches the first, second, and third AI computer models of claim 1 above. Monea further teaches the third AI computer model, wherein processing, the natural language narrative to extract features based on the computer NLP of the natural language narrative is trained through a machine learning process processing to determine the threat level of the code change comprises (classification platform 118, based on the provided NL as disclosed in [0043], determines different types/pattern/feature of malicious activities, e.g. [0013] " where each maliciousness score indicates whether source codes are associated with one or more types (e.g., specific type of calls) of malicious activity. For example, the maliciousness score indicates that the source code is associated with the one or more types of malicious activity or the source code fails to meet the criteria for being associated with the one or more types of malicious activity”; [0043] “The CSM agent 106 may be configured to provide the NL descriptions of the source code to a classification model (e.g., enhanced classification platform 118) trained to generate a first set of maliciousness scores each indicating whether source codes are associated with one or more types of malicious activity. The CSM agent 106 may be configured to generate, using the classification model (e.g., enhanced classification platform 118), a maliciousness score for the source code indicating that the source code is associated with the one or more types of malicious activity”); Note that the third AI computer model is interpreted as the reference’s Classification Platform 118 to generate a threat level, second set of maliciousness scores, each indicating whether the source codes is associated with malicious activity, where the score/ threat level is determined based on the provided NL and processed NL using the classification platform 118.
However, the combination of Monea and Baker does not explicitly teach the listed limitation below. Kim further teaches comparing the code change against a database of known code change patterns based on a semantic similarity matching, wherein the known code change patterns specify at least one of known vulnerabilities or known acceptable modifications ([0566] “Natural language processing analysis such as TF-IDF is performed on an extracted specific pattern (c)”; [0567] “A data set having high similarity is selected among data sets having patterns related to an existing attack identifier (T-ID), and the remaining data sets are filtered (d). In this instance, it is possible to select sample data sets including some or all features of the data sets having patterns related to the attack identifier (T-ID) by comparing with data sets stored in an existing semantic pattern set”); and
determining a threat level based on a degree of matching of the code change to known code change patterns ([0569] “A probability is obtained for each attack identifier (T-ID) by inputting the vectorized N-gram data into the classification model. For example, it is possible to obtain A % as a probability that vectorized data of an N-gram structure is a specific attack identifier (T-ID) T1027, and obtain (100-A) % as a probability that vectorized data of an N-gram structure is an attack identifier T1055”; [0579] “Labeling may be added to the identified attacker or attacker group (Attacker or Group) as described above. This figure illustrates an example in which the attacker TA504 is identified by labeling of the attacker or attacker group (Attacker or Group)”).
Monea, Baker, and Kim analogous in systems that detects source code changes. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea in view of Baker to incorporate the teachings of Kim to implement known change patterns and determining a threat level. Doing so would strengthen identifying changes in code and determine appropriate course of action (Kim [0010] “In connection with the above problems, the conventional method has a problem in that it is difficult to predict a type of cyber threat attack occurring in the near future by such an individual case-focused detection method”).
Regarding the computer program product claim 18, the claim recites similar limitations as the computer method claim 8, therefore, rejected based on the same rationale as claim 8.
Claim(s) 9 and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Monea (US 20250217473A1) in view of Baker (US20250272091A1) and Dalton (US 20230362173 A1).
Regarding claim 9, the combination of Monea and Baker teaches the first, second, and third AI computer models listed in claim 1 listed above. Monea teaches further wherein the threat level of the code change is calculated as [a weighted function of a plurality of scores] capturing different aspects of the code change (classification platform 118, based on the provided NL as disclosed in [0043], determines different types/pattern/feature of malicious activities, e.g. [0013] " where each maliciousness score indicates whether source codes are associated with one or more types (e.g., specific type of calls) of malicious activity. For example, the maliciousness score indicates that the source code is associated with the one or more types of malicious activity or the source code fails to meet the criteria for being associated with the one or more types of malicious activity”).
However, the combination of Monea and Baker does not explicitly teach the limitation listed below. Dalton further teaches a weighted function of a plurality of score ([0066]“ and analyzing, by the at least one electronic computer processor, currently monitored inputs and outputs of the SATCOM network to determine a relative security confidence score of an authenticity of each of said plurality of user terminals of the plurality of user terminals, which may include, e.g., but not limited to, at least one or more of: comparing, by the at least one electronic computer processor, the currently monitored inputs and outputs of the SATCOM network to the at least one logged parameters history for each of the user terminals of the plurality of user terminals to determine the relative security confidence score of the authenticity of each of the plurality of user terminals of the user network based upon at least one of: calculating, by the at least one electronic computer processor, a statistical threat probability of the relative security confidence score of the authenticity of each user terminal of the plurality of user terminals, based upon a mathematical statistical probability weighting of the at least one or more of: the at least one RF signal property metric being monitored; or the at least one network level demodulated signal traffic metric; comparing, by the at least one electronic computer processor, the calculated statistical threat probability of the relative security confidence score of the authenticity of each user terminal of the plurality of user terminals to a predetermined intrusion threat confidence threshold; determining, by the at least one electronic computer processor, if the calculated statistical threat probability of the relative security confidence score of the authenticity of each user terminal of the plurality of user terminals exceeds the predetermined intrusion threat confidence threshold; or generating, by the at least one electronic computer processor, at least one alert, alert, or notification of the intrusion, based on said determining of said (iii)”).
Monea, Baker, and Dalton are analogous in systems that implements machine learning. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea to incorporate the teachings of Dalton to implement a weighted function of scores. Doing so would strengthen identifying changes in code and determine appropriate course of action (Dalton [0008]“Another one of the instructions may include determining, based on the streaming metrics data, whether a confidence score for the identified terminal meets or exceeds a predetermined threshold. Yet another one of the instructions may include generating an alert based upon the evaluation.”).
Regarding the computer program product claim 19, the claim recites similar limitations as the computer method claim 9, therefore, rejected based on the same rationale as claim 9.
Claim(s) 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over Monea (US 20250217473A1) in view of Baker (US20250272091A1) and Dalton (US 20230362173 A1) and Kim (US 20230252145A1).
Regarding claim 10, the combination of Monea, Baker, and Dalton teaches all of the limitations of claim 9 listed above. However, the combination does not explicitly teach the limitation listed below. Kim teaches further the method of claim 9, wherein the plurality of scores comprise at least one of:
an Anomaly Score that indicates how different the code change is from expected code changes;
a Pattern_Match_Score that indicates how much the code change matches known patterns indicative of malicious code ([0491 “As an example for processing the similarity of the training data in this step, when determining an attack identifier or class (T-ID) from high-dimensional data, the TF-IDF technique may be used to select a meaningful feature (pattern). In general, the TF-IDF technique is used to find documents having high similarity in a search engine, and equations for calculating this value are as follows”, see Equation1, Equation2, and Equation3 of reference; [0495] “In an embodiment, a document including a corresponding word may be inferred as an attack identifier (T-ID) based on a characteristic or pattern of a word in block-unit code. Therefore, when the TF-IDF is calculated with respect to a pattern extracted from the block-unit code, a pattern that appears frequently within a specific attack identifier (T-ID) may be extracted, or code having a pattern unrelated to the specific attack identifier (T-ID) may be removed” ), requires a score that represents how well that code change matches a known pattern. The system calculates a degree of match which can be quantified as a score;
a Source_Auth_Score that indicates how reliable a source of the code change 1s; or
a Historical_Deviation_Score that indicates how much the code change deviates from previous historical code changes made to resources.
Monea, Baker, Dalton, and Kim are analogous in systems that detects source code changes. Therefore, it would have been obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to have modified Monea in view of Baker and Dalton to incorporate the teachings of Kim to implement known change patterns and determining a threat level. Doing so would strengthen identifying changes in code and determine appropriate course of action (Kim [0010] “In connection with the above problems, the conventional method has a problem in that it is difficult to predict a type of cyber threat attack occurring in the near future by such an individual case-focused detection method”).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's
disclosure. Bell (US 8375361 B2) discloses a data processing method implementing a source code repository with changes ([Abstract] “A data processing method comprises maintaining a source code repository, checking out a component from the source code repository, checking in a modified version of the component to the source code repository, identifying the changes between the modified version of the component and the original component, applying one or more pattern matching techniques to the identified changes, storing the output of the pattern matching techniques, and displaying the output”).
THIS ACTION IS MADE FINAL. 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 VIVIAN D. HO whose telephone number is (571)272-9957. The examiner can normally be reached M- TH 9:00 - 6:00; F 9:00-1:00.
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, Eleni A. Shiferaw can be reached at (571) 272-3867. 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.
/VIVIAN D HO/Examiner, Art Unit 2497 /ELENI A SHIFERAW/Supervisory Patent Examiner, Art Unit 2497