# US 12,761,883 B2 · Securing software development cycles with artificial intelligence customization > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,761,883 B2 - Document ID: US12761883B2 - Status: Granted - Kind: B2 - Application number: 18/426,627 - Filed: 2024-01-30 - Published: 2025-07-31 - Granted: 2026-10-13 - Earliest claimed: 2024-01-30 - Assignees: CyberArk Software Ltd. - Applicants: CyberArk Software Ltd. - Inventors: Avishay Bar; Erez Waisbard - Web page: https://patentgrep.com/patent/US12761883B2 ## CPC - G06F 8/35: model driven - G06F 40/20: Natural language analysis (semantic analysis of natural language G06F40/30) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Techniques are provided for securing software development cycles with artificial intelligence customization. Operations may include identifying a software generation task; providing the software generation task to a language model; identifying, from the language model, a plurality of queries associated with specific attributes of the software generation task; creating, based on the software generation task and responses to the plurality of queries, a reconstructed software generation task; assigning, based on a machine learning model, one or more security labels to the reconstructed software generation task; determining one or more prioritization scores for one or more security rules based on the one or more security labels; and generating, based on the one or more prioritization scores, at least one security action for the reconstructed software generation task. ## Claims ### Claim 1 A non-transitory computer readable medium including instructions that, when executed by at least one processor, cause the at least one processor to perform operations for interactively enhancing security in software development life cycles, the operations comprising: identifying a software generation task; providing the software generation task to a language model; executing the language model to generate a plurality of queries associated with specific attributes of the software generation task; creating, based on the software generation task and responses to the plurality of queries, a reconstructed software generation task; assigning, based on a machine learning model, one or more security labels to the reconstructed software generation task; determining one or more prioritization scores for one or more security rules based on the one or more security labels; generating, based on the one or more prioritization scores, at least one security action for the reconstructed software generation task; and generating program code to implement the at least one security action. ### Claim 2 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the operations further comprise determining, using the machine learning model, at least one relevancy score for the one or more security labels, wherein the at least one relevancy score indicates a degree of relevancy or prioritization of the one or more security labels to the reconstructed software generation task. ### Claim 3 (depends on 2) The non-transitory computer readable medium of claim 2, wherein assigning the one or more security labels to the reconstructed software generation task is based on the at least one relevancy score. ### Claim 4 (depends on 2) The non-transitory computer readable medium of claim 2, wherein the operations further comprise performing a historical risk analysis for the reconstructed software generation task. ### Claim 5 (depends on 4) The non-transitory computer readable medium of claim 4, wherein the historical risk analysis comprises analyzing historical activity of a user or group associated with the software generation task or the reconstructed software generation task. ### Claim 6 (depends on 4) The non-transitory computer readable medium of claim 4, wherein the operations further comprise assigning one or more scores based on the historical risk analysis. ### Claim 7 (depends on 6) The non-transitory computer readable medium of claim 6, wherein the one or more scores based on the historical risk analysis indicate a number of past instances of a security issue for a user or group associated with the software generation task or the reconstructed software generation task. ### Claim 8 (depends on 6) The non-transitory computer readable medium of claim 6, wherein determining the one or more prioritization scores comprises analyzing, using a model, one or more of: the one or more scores based on the historical risk analysis, industry best practices for security, organization best practices for security, or an input from a user. ### Claim 9 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the software generation task includes at least one of: a software creation task, a software update task, or a software deletion task. ### Claim 10 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the language model comprises a trained language model. ### Claim 11 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the operations further comprise determining the one or more security rules based on the one or more security labels. ### Claim 12 (depends on 11) The non-transitory computer readable medium of claim 11, wherein determining the one or more security rules based on the one or more security labels comprises mapping each of a plurality of security labels to a corresponding set of one or more security rules. ### Claim 13 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the at least one security action comprises a plurality of security actions presented in accordance with a prioritization based on the one or more prioritization scores. ### Claim 14 (depends on 1) The non-transitory computer readable medium of claim 1, wherein generating the at least one security action comprises: determining whether each of the one or more prioritization scores satisfies a threshold; and defining, for each of the one or more security rules whose prioritization score satisfies the threshold, one or more security actions. ### Claim 15 (depends on 1) The non-transitory computer readable medium of claim 1, wherein the machine learning model comprises the language model, another language model, or a classification model. ### Claim 16 A computer-implemented method for interactively enhancing security in software development life cycles, the method comprising: identifying a software generation task; providing the software generation task to a language model; executing the language model to generate a plurality of queries associated with specific attributes of the software generation task; creating, based on the software generation task and responses to the plurality of queries, a reconstructed software generation task; assigning, based on a machine learning model, one or more security labels to the reconstructed software generation task; determining one or more prioritization scores for one or more security rules based on the one or more security labels; generating, based on the one or more prioritization scores, at least one security action for the reconstructed software generation task; and generating program code to implement the at least one security action. ### Claim 17 (depends on 16) The computer-implemented method of claim 16, wherein the reconstructed software generation task comprises a plurality of requirements associated with software development. ### Claim 18 (depends on 16) The computer-implemented method of claim 16, wherein the one or more prioritization scores are based on at least one of: a task classification score or a historical risk analysis score. ### Claim 19 (depends on 16) The computer-implemented method of claim 16, further comprising: implementing the at least one security action based on a code generation engine. ### Claim 20 (depends on 16) The computer-implemented method of claim 16, further comprising: based on user feedback provided by a software developer or a security reviewer, training or updating at least one of: the language model, the machine learning model, a model configured to map security labels to security rules, a model configured to determine the one or more prioritization scores, or a model configured to conduct a historical risk analysis.