Automated validation of suspicious activity report narratives using generative artificial intelligence

Patent metadata

Field	Value
Publication number	US-12536544-B2
Application number	US-202418422858-A
Country	US
Kind code	B2
Filing date	Jan 25, 2024
Priority date	Jan 25, 2024
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

An autonomous fraud/AML reporting system and methods are provided that are configured to automate validations of SAR narratives using a generative AI service by an automated SAR narrative system. The system includes a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform narrative validation operations which include receiving a SAR narrative for a SAR, loading a prompt template associated with validating the SAR narrative by the generative AI service, injecting the narrative into the prompt templates, and generating and storing the validation based on the comparing.

First claim

What is claimed is:

1 . An automated suspicious activity report (SAR) narrative system configured to automate validations of SAR narratives using a generative artificial intelligence (AI) service, the automated SAR narrative system comprising:
a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform narrative validation operations which comprise:
receiving a SAR narrative for a SAR, wherein the SAR narrative includes an explanation of SAR data included in the SAR;
parsing a JavaScript Object Notation (JSON) object for the SAR narrative for one or more entities present in the explanation of the SAR data using a software library;
generating a plurality of new JSON objects for each of the one or more entities based on parsing the JSON object for the SAR narrative;
loading a prompt template into a local memory of the automated SAR narrative system, wherein the prompt template is associated with validating the SAR narrative by the generative AI service, and wherein the prompt template includes instructions to the generative AI service that cause a validation of the SAR narrative for the SAR and details of the SAR that are used for the validation of the SAR narrative;
extracting, from the plurality of new JSON objects and the SAR narrative, narrative data corresponding to one or more of the details;
generating a prompt to the generative AI service based on the extracted narrative data and the prompt template, wherein the generating the prompt comprises:
identifying input fields of the prompt template that correspond to the extracted narrative data and the one or more of the details,
inserting the extracted narrative data into the input fields, and
creating the prompt based on the prompt template having the extracted narrative data inserted to the input fields;

comparing, via one or more calls to the generative AI service, the SAR narrative to the SAR data included in the SAR using a prompting strategy to the generative AI service and the prompt, wherein the prompting strategy uses one or more text commands made to the generative AI service through the one or more calls; and
generating and storing the validation of the SAR narrative based on the comparing, wherein the storing comprises adding the validation to a data container including the SAR and the SAR narrative.

2 . The automated SAR narrative system of claim 1 , wherein prompting strategy utilizes a question-and-answer-based validation technique that questions the generative AI service with a plurality of questions from the prompt while passing the SAR narrative to the generative AI service.

3 . The automated SAR narrative system of claim 1 , wherein the prompting strategy utilizes a large language model (LLM)-as-a-reviewer technique that prompts an LLM of the generative AI service to review the SAR narrative and identify any mismatches, any gaps, or a combination thereof between the extracted narrative data and the SAR data in the SAR.

4 . The automated SAR narrative system of claim 1 , wherein the prompting strategy utilizes the instructions to prompt the generative AI service to return a JSON format data structure having SAR data values corresponding to a plurality of SAR fields for the SAR, and wherein the SAR data values in the JSON format data structure are correlated with the SAR data for use with the comparing.

5 . The automated SAR narrative system of claim 1 , wherein the comparing is performed by each section of a plurality of sections of the SAR narrative.

6 . The automated SAR narrative system of claim 1 , wherein the comparing is performed for each entity of a plurality of entities present in the SAR narrative.

7 . The automated SAR narrative system of claim 1 , wherein the prompt template is one of at least two prompt templates for use with the validation of the SAR narrative, and wherein each of the at least two prompt templates are used for creating corresponding prompts for the validation that are run in parallel with the generative AI service using a multithreading-based processing job.

8 . The automated SAR narrative system of claim 7 , wherein the prompting strategy causes the comparing to be performed through a plurality of individual calls to the generative AI service in a specified order designated by the instructions for the one of the at least two prompt templates being run.

9 . The automated SAR narrative system of claim 1 , wherein the generative AI service comprises at least one generative AI model including at least an LLM, and wherein the generative AI service provides a conversational AI that processes input conversational text corresponding to the prompt and provides output conversational text having the validation of the SAR narrative.

10 . A method to automate validations of suspicious activity report (SAR) narratives using a generative artificial intelligence (AI) service for an automated SAR narrative system, the method comprising:
receiving a SAR narrative for a SAR, wherein the SAR narrative includes an explanation of SAR data included in the SAR; parsing a JavaScript Object Notation (JSON) object for the SAR narrative for one or more entities present in the explanation of the SAR data using a software library; generating a plurality of new JSON objects for each of the one or more entities based on parsing the JSON object for the SAR narrative; loading a prompt template into a local memory of the automated SAR narrative system, wherein the prompt template is associated with validating the SAR narrative by the generative AI service, and wherein the prompt template includes instructions to the generative AI service that cause a validation of the SAR narrative for the SAR and details of the SAR that are used for the validation of the SAR narrative; extracting, from the plurality of new JSON objects and the SAR narrative, narrative data corresponding to one or more of the details; generating a prompt to the generative AI service based on the extracted narrative data and the prompt template, wherein the generating the prompt comprises:
identifying input fields of the prompt template that correspond to the extracted narrative data and the one or more of the details,
inserting the extracted narrative data into the input fields, and
creating the prompt based on the prompt template having the extracted narrative data inserted to the input fields;
comparing, via one or more calls to the generative AI service, the SAR narrative to the SAR data included in the SAR using a prompting strategy to the generative AI service and the prompt, wherein the prompting strategy uses one or more text commands made to the generative AI service through the one or more calls; and generating and storing the validation of the SAR narrative based on the comparing, wherein the storing comprises adding the validation to a data container including the SAR and the SAR narrative.

11 . The method of claim 10 , wherein prompting strategy utilizes a question-and-answer-based validation technique that questions the generative AI service with a plurality of questions from the prompt while passing the SAR narrative to the generative AI service.

12 . The method of claim 10 , wherein the prompting strategy utilizes a large language model (LLM)-as-a-reviewer technique that prompts an LLM of the generative AI service to review the SAR narrative and identify any mismatches, any gaps, or a combination thereof between the extracted narrative data and the SAR data in the SAR.

13 . The method of cla

Assignees

• Actimize Ltd

Inventors

• Bathula Kiran Kumar

Classifications

• G06Q20/4016 Primary

  involving fraud or risk level assessment in transaction processing · CPC title