Verification method, system, storage medium and device for rpa script

Abstract

The application discloses a kind of verification method, system, storage medium and equipment of RPA script, belong to artificial intelligence technical field.AI Agent will user intent split into execution step;RPA script designer generates an RPA script to each execution step;RPA script executor generates static ZKP according to the version information and script content of RPA script, AI Agent utilizes static ZKP to verify the integrity of RPA script;For each RPA script, RPA script executor records execution data in the execution process of RPA script, generates dynamic ZKP according to execution data, AI Agent utilizes dynamic ZKP to verify the correctness of RPA script execution process;AI Agent utilizes all dynamic ZKP to verify the consistency of the complete execution flow of all RPA scripts.The application can solve the problem that RPA script is tamperable, execution process is not verifiable, privacy disclosure.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, system, storage medium and device for verifying RPA scripts. Background Technology

[0002] Against the backdrop of rapidly evolving enterprise automation and intelligence demands, Robotic Process Automation (RPA) is widely used in scenarios such as data acquisition, process-driven operations, and system interaction. Meanwhile, Artificial Intelligence Agents (AI Agents) are increasingly becoming the mainstream approach for building complex business collaboration platforms. The combination of these two technologies can perfectly achieve the fusion of automation and intelligence.

[0003] However, combining AI Agents with RPA still faces the following problems:

[0004] (1) Risk of script tampering: If the RPA script is tampered with before execution, it may lead to misoperation or malicious behavior, and there is a lack of effective integrity verification mechanism;

[0005] (2) Lack of execution consistency: It cannot be proven whether the RPA execution result conforms to the user's intention of the original AIAgent, which affects the credibility and compliance of the AIAgent;

[0006] (3) Privacy protection needs are prominent: While ensuring verifiability, it is also necessary to protect execution data and process logic from being leaked.

[0007] Therefore, there is an urgent need for a secure, reliable, composable and verifiable RPA execution behavior assurance mechanism for AIAgent systems, which takes into account both data privacy and operational verifiability. Summary of the Invention

[0008] This application provides a method, system, storage medium, and device for verifying RPA scripts, addressing the risks of RPA script tampering, the inability to prove that the RPA execution result conforms to the user intent of the original AIAgent, and the risk of leakage of execution data and process logic when combining AIAgent with RPA. The technical solution is as follows:

[0009] According to a first aspect of this application, an RPA script verification method is provided for use in an RPA script verification system, the verification system including an AI Agent and RPA, the RPA including an RPA script designer and an RPA script executor, the method comprising:

[0010] The AI ​​Agent breaks down the acquired user intent into multiple execution steps;

[0011] The RPA script designer generates an RPA script for each execution step;

[0012] The RPA script executor generates static zero-knowledge proofs (ZKPs) based on the version information and script content of the RPA script.

[0013] The AI ​​Agent uses the static ZKP to verify the integrity of the RPA script;

[0014] For each RPA script, the RPA script executor records execution data during the execution of the RPA script and generates a dynamic ZKP based on the execution data;

[0015] The AI ​​Agent uses the dynamic ZKP to verify the correctness of the RPA script execution process;

[0016] The AI ​​Agent utilizes all dynamic ZKP to verify the consistency of the complete execution flow of all RPA scripts.

[0017] In one possible implementation, the RPA script executor generates a static ZKP based on the RPA script's version information and script content, including:

[0018] The RPA script executor performs a hash operation on the script content of the RPA script to obtain the header hash value;

[0019] The RPA script executor performs a hash operation on the sum of the RPA script version information and the header hash to obtain the tail hash value;

[0020] The RPA script executor generates the circuit constraints of the static ZKP based on the header hash value and the tail hash value.

[0021] In one possible implementation, the AI ​​Agent verifies the integrity of the RPA script using the static ZKP, including:

[0022] The AI ​​Agent verifies whether the hash value of the script content is the same as the header hash value based on the circuit constraints of the static ZKP, and verifies whether the hash value of the sum of the version information and the header hash is the same as the tail hash value.

[0023] If both are true, then the AI ​​Agent determines that the RPA script is complete.

[0024] In one possible implementation, when the execution data includes a script number, script input, and script output, the RPA script executor generates a dynamic ZKP based on the execution data, including:

[0025] The RPA script executor performs a hash operation on the script input to obtain the input hash value;

[0026] The RPA script executor performs a hash operation on the script output to obtain the output hash value;

[0027] The RPA script executor generates an output identifier based on the script output, and the output identifier indicates whether the script output is valid or invalid.

[0028] The RPA script executor performs a hash operation on the sum of the script input and the script output, adds the resulting input-output hash value to the previous hash value to obtain the current hash value, where the previous hash value is the hash value calculated for the previous RPA script;

[0029] The RPA script executor generates the circuit constraints of the dynamic ZKP based on the input hash value, the output hash value, the output identifier, and the current input-output hash value.

[0030] In one possible implementation, the AI ​​Agent utilizes the dynamic ZKP to verify the correctness of the RPA script execution process, including:

[0031] The AI ​​Agent verifies whether the hash value of the script input is equal to the input hash value based on the circuit constraints of the dynamic ZKP, verifies whether the hash value of the script output is equal to the output hash value, verifies whether the output identifier is generated based on the verification of the script output, and verifies whether the sum of the input-output hash value and the previous hash value is equal to the current hash value. The input-output hash value is obtained by performing a hash operation on the sum of the script input and the script output.

[0032] If both are true, then the AI ​​Agent determines that the execution process of the RPA script is correct.

[0033] In one possible implementation, the execution data further includes an execution timestamp, in which case the AI ​​Agent uses the dynamic ZKP to verify the correctness of the RPA script execution process, including:

[0034] The AI ​​Agent verifies whether the RPA script is executed within a predetermined time range based on the execution timestamp;

[0035] If so, the AI ​​Agent determines that the execution process of the RPA script is correct.

[0036] In one possible implementation, the AI ​​Agent utilizes all dynamic ZKP to verify the consistency of the complete execution flow of all RPA scripts, including:

[0037] The AI ​​Agent reads the current hash value from the i-th dynamic ZKP and the previous hash value from the (i+1)-th dynamic ZKP, and verifies whether the current hash value is equal to the previous hash value, where i is a positive integer;

[0038] The AI ​​Agent reads the previous hash value, input hash value, and output hash value from the i-th dynamic ZKP, reads the previous hash value from the (i+1)-th dynamic ZKP, and verifies whether the hash value of the sum of the input hash value and the output hash value, when added to the i-th previous hash value, is equal to the (i+1)-th previous hash value.

[0039] If both are true, then the AI ​​Agent determines that the execution flow of all RPA scripts is complete.

[0040] According to a second aspect of this application, an RPA script verification system is provided, the verification system including AIAgent and RPA, the RPA including an RPA script designer and an RPA script executor;

[0041] The AI ​​Agent is used to break down the acquired user intent into multiple execution steps;

[0042] The RPA script designer is used to generate an RPA script for each execution step;

[0043] The RPA script executor is used to generate static zero-knowledge proofs (ZKPs) based on the version information and script content of the RPA script.

[0044] The AI ​​Agent is also used to verify the integrity of the RPA script using the static ZKP;

[0045] The RPA script executor is also used to record execution data during the execution of each RPA script and generate a dynamic ZKP based on the execution data.

[0046] The AI ​​Agent is also used to verify the correctness of the RPA script execution process using the dynamic ZKP;

[0047] The AI ​​Agent is also used to verify the consistency of the complete execution flow of all RPA scripts using all dynamic ZKP.

[0048] According to a third aspect of this application, a computer-readable storage medium is provided, wherein at least one instruction is stored therein, the at least one instruction being loaded and executed by a processor to implement the verification method of the RPA script as described above.

[0049] According to a fourth aspect of this application, a computer device is provided, the computer device including the verification system for the above-described RPA script.

[0050] The beneficial effects of the technical solution provided in this application include at least the following:

[0051] The RPA script executor generates static ZKPs based on the RPA script's version information and content. The AI ​​Agent uses these static ZKPs to verify the integrity of the RPA script. The RPA script executor also generates dynamic ZKPs based on the RPA script's execution data. The AI ​​Agent uses these dynamic ZKPs to verify the correctness of the RPA script execution process. By utilizing all dynamic ZKPs, the consistency of the complete execution flow of all RPA scripts is verified. This dual ZKP mechanism solves three major problems: RPA script tampering, unverifiable execution process, and privacy leakage. It improves the traceability of task execution, provides trusted combined proofs across AI Agent execution flows, and balances data protection and execution verification, making it suitable for scenarios with high security requirements. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0053] Figure 1 This is a flowchart of an RPA script verification method provided in one embodiment of this application;

[0054] Figure 2 This is a schematic diagram of user intent, intent constraint logic rule annotation, and RPA script provided in one embodiment of this application;

[0055] Figure 3 This is a schematic diagram of static ZKP and dynamic ZKP provided in one embodiment of this application;

[0056] Figure 4 This is a structural block diagram of an RPA script verification system provided in one embodiment of this application. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0058] This application involves static zero-knowledge proofs (ZKP) and dynamic ZKP. Static ZKP is generated based on the content of the ZKP script and is used to verify the integrity of the RPA script, ensuring that the RPA script has not been tampered with. Dynamic ZKP is generated based on the execution data collected during the execution of the RPA script and is used to verify the correctness of the execution process of the RPA script, ensuring that it conforms to the user's intent, while also ensuring that the execution data and execution logic are not leaked. Dynamic ZKP is also used to verify the consistency of the task flow, ensuring the consistency of the execution process of the RPA script before and after execution.

[0059] like Figure 1 The diagram illustrates a flowchart of an RPA script verification method according to an embodiment of this application. This RPA script verification method can be applied to an RPA script verification system, which includes an AI Agent and RPA. The RPA includes an RPA script designer and an RPA script executor. The RPA script verification method may include:

[0060] Step 101: The AI ​​Agent breaks down the acquired user intent into multiple execution steps.

[0061] User intent refers to the purpose that a user wants to achieve using AI Agent and RPA. For example, in an application scenario where AI Agent assists in processing corporate financial expense reimbursement processes, the user intent could be "Please help me process employee expense reports in batches."

[0062] Based on an understanding of the user's intent, the AI ​​agent meticulously plans the user's tasks and breaks them down into multiple execution steps.

[0063] Taking the above application scenario as an example, the AI ​​Agent's step-by-step planning for understanding user intent includes:

[0064] Step 1: Obtain the list of expense reports to be processed;

[0065] Step 2: Verify whether the invoice information is compliant;

[0066] Step 3: Automatically fill in the system based on employee information and budget;

[0067] Step 4: Submit the reimbursement request to the approval system.

[0068] Since the execution steps planned by the AI ​​Agent based on user intent are a series of abstract, semantically driven task flows, while RPA scripts are a series of concrete atomic action task flows, a certain semantic relationship is needed between the two to achieve matching between the abstract intent of the AI ​​Agent and the execution steps of the RPA script. Therefore, the AI ​​Agent needs to annotate the multiple execution steps broken down from the user intent with intent constraint logic rules, using these rules as a bridge between the AI ​​Agent and the RPA script for intent recognition and rule constraints. This identifies the semantic intent of each RPA script and imposes rule constraints on the implementation of the RPA script.

[0069] The metadata for intention constraint logic rule annotations includes:

[0070] Intent_id: Intent number. Each intent number corresponds to an execution step.

[0071] Desp: A semantic description of the RPA script, explaining the execution intent of the RPA script.

[0072] Input: Determines the types of inputs that the RPA script can accept, the input structure, and specific restrictions, such as "the invoice format must meet a certain type of format validation".

[0073] Output: Specify the data types, data structures, and specific limitations that the RPA script can output, such as the validation result being either valid or invalid.

[0074] Figure 2 This illustrates the correspondence between user intents and intent constraint logic rule annotations:

[0075] (1) Step 1 is "Get the list of expense reports to be processed".

[0076] The Desp corresponding to Intent No. 1 is "Load Expense Reports to be Processed", the Input is "Expense Report Form", and the Output is "data".

[0077] (2) Step 2 is to “verify whether the invoice information is compliant”.

[0078] The Desp corresponding to Intent No. 2 is "Read invoice content and verify invoice compliance", the Input is "data", and the Output is "valid / invalid".

[0079] (3) Step 3 is to “automatically fill in the system based on employee information and budget”.

[0080] The Desp corresponding to Intent No. 3 is "Compare reimbursement results and fill in the reimbursement process based on employee information and budget". The Input includes employee_id and data, and the Output is result.

[0081] (4) Step 4 is to "submit the reimbursement request to the approval system".

[0082] The Desp corresponding to Intent No. 4 is "Submit expense report and synchronize it to the approval system", with Input being result and Output being True / False.

[0083] Step 102: The RPA script designer generates an RPA script for each execution step.

[0084] The RPA script designer can transform each execution step into a series of atomic actions based on intent constraint logic rules, such as mouse clicks, data input, file copying, and numerical calculations, to achieve a fixed task flow, thereby obtaining an RPA script with operation flows such as input operations, processing operations, and output operations.

[0085] Figure 2 This illustrates the correspondence between user intents, intent constraint logic rule annotations, and RPA scripts:

[0086] (1) Step 1 is "Get the list of expense reports to be processed".

[0087] RPA_script_1: read_csv(“Expense Report Form”), data=output(employ_id, expense details).

[0088] (2) Step 2 is to “verify whether the invoice information is compliant”.

[0089] RPA_script_2:output=verify (invoice compliance).

[0090] (3) Step 3 is to “automatically fill in the system based on employee information and budget”.

[0091] RPA_script_3: result=compare(reimbursement amount, budget).

[0092] (4) Step 4 is to "submit the reimbursement request to the approval system";

[0093] RPA_script_4: result=commit(reimbursement information).

[0094] Step 103: The RPA script executor generates a static ZKP based on the RPA script version information and script content.

[0095] For each RPA script, the RPA script executor can insert a header hash and a footer hash into the RPA script to record a summary of the original content of the RPA script and generate a static ZKP. Before and after each execution of the RPA script, the AI ​​Agent compares and analyzes the file hash of the current RPA script to ensure that the RPA script has not been tampered with.

[0096] Specifically, the RPA script executor generates static ZKP based on the RPA script version information and script content, which may include:

[0097] (1) The RPA script executor performs a hash operation on the script content of the RPA script to obtain the header hash value.

[0098] Based on the script content of the RPA script, the RPA script executor calculates its hash: script_hash = Hash(script_code); and inserts script_hash into the header of the RPA script, that is, the header hash value header_hash = script_hash.

[0099] (2) The RPA script executor performs a hash operation on the sum of the RPA script version information and the header hash to obtain the tail hash value.

[0100] Based on the RPA script's version information and script content, the RPA script executor calculates its hash: footer_hash = Hash(script_version + script_hash); and inserts footer_hash at the end of the RPA script, i.e., the tail hash value footer_hash = footer_hash.

[0101] (3) The RPA script executor generates static ZKP circuit constraints based on the header hash value and the tail hash value.

[0102] The RPA script executor generates static ZKP based on RPA script hashes:

[0103] Private variables: script_code, script_version

[0104] Public variables: header_hash, footer_hash

[0105] Circuit constraints of static ZKP:

[0106] (1) C1: header_hash==hash(script_code);

[0107] (2) C2: footer_hash==hash (script_version + header_hash).

[0108] Step 104: The AI ​​Agent uses static ZKP to verify the integrity of the RPA script.

[0109] Specifically, the AI ​​Agent uses static ZKP to verify the integrity of the RPA script, which may include: the AI ​​Agent verifying whether the hash value of the script content is the same as the hash value of the header based on the circuit constraints of the static ZKP, and verifying whether the hash value of the sum of the version information and the header hash is the same as the hash value of the tail; if both are true, the AI ​​Agent determines that the RPA script is complete; otherwise, the AI ​​Agent determines that the RPA script is incomplete.

[0110] That is, the AI ​​Agent verifies the RPA script based on the public variables of static ZKP (such as head_hash and footer_hash of the RPA script) and the hash calculation process of circuit constraints, and can confirm that the RPA script has not been tampered with.

[0111] Step 105: For each RPA script, the RPA script executor records execution data during the execution of the RPA script and generates a dynamic ZKP based on the execution data.

[0112] The execution process of each RPA script can generate a verifiable dynamic ZKP, which is convenient for the AI ​​Agent to further verify and ensure that the actual execution path of each RPA script is consistent with the AI ​​Agent's preset path and does not deviate from the established business objectives.

[0113] Specifically, when the execution data includes a script number, script input, and script output, the RPA script executor generates a dynamic ZKP based on the execution data. This can include: the RPA script executor performing a hash operation on the script input to obtain an input hash value; the RPA script executor performing a hash operation on the script output to obtain an output hash value; the RPA script executor generating an output identifier based on the script output, which indicates whether the script output is valid or invalid; the RPA script executor performing a hash operation on the sum of the script input and script output, adding the resulting input-output hash value to the previous hash value to obtain the current hash value, where the previous hash value is the hash value calculated for the previous RPA script; and the RPA script executor generating circuit constraints for the dynamic ZKP based on the input hash value, output hash value, output identifier, and current input-output hash value.

[0114] Taking "verifying whether invoice information is compliant" as an example, the execution data is as follows:

[0115] RPA_id: The number of the RPA script.

[0116] RPA_input: The input to the RPA script, which in this case is the invoice content.

[0117] RPA_output: The output of the RPA script, here representing the validation result, such as valid or invalid.

[0118] RPA_timestamp: The execution timestamp of the RPA script.

[0119] The dynamic ZKP generated based on the above execution data is as follows:

[0120] Private variables are shown in Table 1:

[0121] Table 1

[0122]

[0123] The public variables are shown in Table 2:

[0124] Table 2

[0125]

[0126] The circuit constraints of dynamic ZKP are:

[0127] (1) C1:RPA_input_hash==hash(RPA_input);

[0128] (2) C2: RPA_output_hash==hash (RPA_output);

[0129] (3) C3: Prove that RPA_output_flag is generated based on the verification of RPA_output;

[0130] (4) C4: Current_hash==prev_hash+ hash (RPA_input + PRA_output).

[0131] Step 106: The AI ​​Agent uses dynamic ZKP to verify the correctness of the RPA script execution process.

[0132] Specifically, the AI ​​Agent uses dynamic ZKP to verify the correctness of the RPA script execution process. This can include: verifying whether the hash value of the script input is equal to the input hash value based on the circuit constraints of the dynamic ZKP; verifying whether the hash value of the script output is equal to the output hash value; verifying whether the output identifier is generated based on the script output verification; and verifying whether the sum of the input / output hash value and the previous hash value is equal to the current hash value. The input / output hash value is obtained by hashing the sum of the script input and script output. If all of these conditions are met, the AI ​​Agent determines that the RPA script execution process is correct; otherwise, the AI ​​Agent determines that the RPA script execution process is incorrect.

[0133] That is, the AI ​​Agent verifies the correctness of the RPA script's output by comparing the public variables of the dynamic ZKP (such as the input / output hash values ​​of the RPA script) with the dynamic ZKP generated during the execution of the RPA script.

[0134] Optionally, the execution process can also be verified from the perspective of the execution time of the RPA script. In this case, the execution data also includes the execution timestamp. The AI ​​Agent uses dynamic ZKP to verify the correctness of the RPA script execution process, which may include: the AI ​​Agent verifying whether the RPA script is executed within the predetermined time range based on the execution timestamp; if so, the AI ​​Agent determines that the RPA script execution process is correct; otherwise, the AI ​​Agent determines that the RPA script execution process is incorrect.

[0135] That is, in addition to verifying the execution process of the RPA script based on the circuit constraints of the dynamic ZKP, the AI ​​Agent also verifies the execution timestamp of the RPA script to determine whether the RPA script is executed within a certain time range.

[0136] Step 107: The AI ​​Agent uses all dynamic ZKP to verify the consistency of the complete execution flow of all RPA scripts.

[0137] Specifically, the AI ​​Agent utilizes all dynamic ZKP to verify the consistency of the complete execution flow of all RPA scripts, which may include:

[0138] (1) The AI ​​Agent reads the current hash value from the i-th dynamic ZKP and the previous hash value from the (i+1)-th dynamic ZKP, and verifies whether the current hash value is equal to the previous hash value, where i is a positive integer.

[0139] That is, the AI ​​Agent verifies the RPA_id. {i+1} .prev_hash==PRA_id {i} .current_hash.

[0140] (2) The AI ​​Agent reads the previous hash value, input hash value and output hash value from the i-th dynamic ZKP, reads the previous hash value from the (i+1)-th dynamic ZKP, and verifies whether the hash value of the sum of the input hash value and the output hash value is equal to the (i+1)-th previous hash value after being added to the i-th previous hash value.

[0141] That is, the AI ​​Agent verifies the RPA_id. {i+1} .prev_hash==RPA_id {i}. prev_hash+Hash(PRA_id) {i} .input_hash+ PRA_id {i} .output_hash).

[0142] (3) If all are true, then the AI ​​Agent determines that the execution flow of all RPA scripts is complete.

[0143] (4) Otherwise, the AI ​​Agent determines that the execution flow of all RPA scripts is incomplete.

[0144] Figure 3 The diagram illustrates the combined use of AI Agent and RPA. The AI ​​Agent first plans the steps based on the user's intent, then annotates the intent constraint logic rules to schedule the RPA script designer to generate the RPA script. The RPA script executor generates a header hash (head_hash) at the beginning of the RPA script and a footer hash (footer_hash) at the end, resulting in a static ZKP. The RPA script executor generates a dynamic ZKP during the execution of the RPA script. The AI ​​Agent verifies both the static and dynamic ZKPs.

[0145] In summary, the RPA script verification method provided in this application involves the RPA script executor generating a static ZKP based on the RPA script version information and script content, and using the static ZKP to verify the integrity of the RPA script; the RPA script executor generating a dynamic ZKP based on the execution data of the RPA script, and using the dynamic ZKP to verify the correctness of the RPA script execution process; and using all dynamic ZKPs to verify the consistency of the complete execution flow of all RPA scripts. This dual ZKP mechanism solves the three major problems of RPA script tampering, unverifiable execution process, and privacy leakage. It improves the traceability of task execution, provides trusted combined proofs of execution processes across AI Agents, and balances data protection and execution verification, making it suitable for scenarios with high security requirements.

[0146] like Figure 4As shown, it illustrates a structural block diagram of an RPA script verification system provided in an embodiment of this application. The RPA script verification system may include: AI Agent 410 and RPA 420. RPA 420 includes an RPA script designer 421 and an RPA script executor 422.

[0147] AI Agent410 is used to break down the acquired user intent into multiple execution steps;

[0148] RPA script designer 421, used to generate an RPA script for each execution step;

[0149] RPA script executor 422 is used to generate static ZKP based on the version information and script content of the RPA script;

[0150] AI Agent410 is also used to verify the integrity of RPA scripts using static ZKP.

[0151] The RPA script executor 422 is also used to record execution data during the execution of each RPA script.

[0152] AI Agent410 is also used to generate dynamic ZKPs based on execution data, and to use dynamic ZKPs to verify the correctness of the RPA script execution process;

[0153] AI Agent410 is also used to verify the consistency of the entire execution flow of all RPA scripts using all dynamic ZKP.

[0154] In an optional embodiment, the RPA script executor 422 is further configured to:

[0155] Perform a hash operation on the content of the RPA script to obtain the header hash value;

[0156] Perform a hash operation on the sum of the RPA script's version information and header hash to obtain the tail hash value;

[0157] The circuit constraints of the static ZKP are generated based on the head hash value and the tail hash value.

[0158] In an optional embodiment, AI Agent 410 is also used for:

[0159] Verify whether the hash value of the script content is the same as the hash value of the header based on the circuit constraints of static ZKP, and verify whether the hash value of the sum of the version information and the header hash is the same as the hash value of the tail.

[0160] If both are true, then the RPA script is complete.

[0161] In an optional embodiment, when the execution data includes a script number, script input, and script output, the RPA script executor 422 is further configured to:

[0162] Perform a hash operation on the script input to obtain the input hash value;

[0163] Perform a hash operation on the script output to obtain the output hash value;

[0164] Generate an output identifier based on the script output; the output identifier indicates whether the script output is valid or invalid.

[0165] Perform a hash operation on the sum of the script input and script output, add the resulting input / output hash value to the previous hash value to obtain the current hash value. The previous hash value is the hash value calculated for the previous RPA script.

[0166] Dynamic ZKP circuit constraints are generated based on the input hash value, output hash value, output identifier, and current input and output hash values.

[0167] In an optional embodiment, AI Agent 410 is also used for:

[0168] Based on the circuit constraints of dynamic ZKP, verify whether the hash value of the input script is equal to the input hash value, verify whether the hash value of the output script is equal to the output hash value, verify whether the output identifier is generated based on the verification of the script output, verify whether the sum of the input and output hash values ​​and the previous hash value is equal to the current hash value, and the input and output hash values ​​are obtained by performing a hash operation on the sum of the script input and the script output;

[0169] If both are true, then the execution process of the RPA script is correct.

[0170] In an optional embodiment, the execution data also includes an execution timestamp, then AI Agent 410 is further used for:

[0171] Verify whether the RPA script was executed within the predetermined time range based on the execution timestamp;

[0172] If so, then the execution process of the RPA script is confirmed to be correct.

[0173] In an optional embodiment, AI Agent 410 is also used for:

[0174] Read the current hash value from the i-th dynamic ZKP, read the previous hash value from the (i+1)-th dynamic ZKP, and verify whether the current hash value is equal to the previous hash value, where i is a positive integer;

[0175] Read the previous hash value, input hash value, and output hash value from the i-th dynamic ZKP. Read the previous hash value from the (i+1)-th dynamic ZKP. Verify whether the hash value of the sum of the input hash value and the output hash value is equal to the (i+1)-th previous hash value after being added to the i-th previous hash value.

[0176] If both are true, then the execution flow of all RPA scripts is complete.

[0177] In summary, the RPA script verification system provided in this application embodiment involves the RPA script executor generating a static ZKP based on the RPA script version information and script content, and using the static ZKP to verify the integrity of the RPA script; the RPA script executor generating a dynamic ZKP based on the execution data of the RPA script, and using the dynamic ZKP to verify the correctness of the RPA script execution process; and using all dynamic ZKPs to verify the consistency of the complete execution flow of all RPA scripts. Thus, through a dual ZKP mechanism, the system solves the three major problems of RPA script tampering, unverifiable execution process, and privacy leakage. It improves the traceability of task execution, provides trusted combined proofs of execution processes across AI Agents, and balances data protection and execution verification, making it suitable for scenarios with high security requirements.

[0178] One embodiment of this application provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the RPA script verification method as described above.

[0179] One embodiment of this application provides a computer device that includes a verification system for any of the RPA scripts described above.

[0180] It should be noted that the RPA script verification system provided in the above embodiments is only illustrated by the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the RPA script verification system can be divided into different functional modules to complete all or part of the functions described above. Furthermore, the RPA script verification system and the RPA script verification method embodiments provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0181] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0182] The above description is not intended to limit the embodiments of this application. Any adjustments, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of the embodiments of this application.

Claims

1. A method for verifying an RPA script, the method comprising: In a verification system for an RPA script, the verification system comprising an AI Agent and an RPA, the RPA comprising an RPA script designer and an RPA script executor, the method comprising: the AI Agent splits the obtained user intent into multiple execution steps; the RPA script designer generates an RPA script for each execution step; the RPA script executor generates a static zero-knowledge proof (ZKP) according to the version information and script content of the RPA script; the AI Agent verifies the integrity of the RPA script using the static ZKP; for each RPA script, the RPA script executor records execution data during the execution of the RPA script, and generates a dynamic ZKP according to the execution data; the AI Agent verifies the correctness of the RPA script execution process using the dynamic ZKP; the AI Agent verifies the consistency of the complete execution process of all RPA scripts using all dynamic ZKPs. 2.The method of claim 1, wherein, The RPA script executor generates a static ZKP according to the version information and script content of the RPA script, comprising: the RPA script executor performs a hash operation on the script content of the RPA script to obtain a header hash value; the RPA script executor performs a hash operation on the sum of the version information and the header hash of the RPA script to obtain a tail hash value; the RPA script executor generates circuit constraints of the static ZKP according to the header hash value and the tail hash value. 3.The method of claim 2, wherein, The AI Agent verifies the integrity of the RPA script using the static ZKP, comprising: the AI Agent verifies whether the hash value of the script content is the same as the header hash value according to the circuit constraints of the static ZKP, and verifies whether the hash value of the sum of the version information and the header hash is the same as the tail hash value; if both are true, the AI Agent determines that the RPA script is complete. 4.The method of claim 1, wherein, When the execution data includes script number, script input and script output, the RPA script executor generates a dynamic ZKP according to the execution data, comprising: the RPA script executor performs a hash operation on the script input to obtain an input hash value; the RPA script executor performs a hash operation on the script output to obtain an output hash value; the RPA script executor generates an output identifier according to the script output, the output identifier indicating that the script output is valid or invalid; the RPA script executor performs a hash operation on the sum of the script input and the script output, adds the obtained input-output hash value to the previous hash value to obtain a current hash value, the previous hash value being a hash value calculated for the previous RPA script; the RPA script executor generates circuit constraints of the dynamic ZKP according to the input hash value, the output hash value, the output identifier and the current hash value. 5.The method of claim 4, wherein, The AI Agent verifies the correctness of the RPA script execution process using the dynamic ZKP, comprising: The AI Agent verifies whether the hash value of the script input is equal to the input hash value according to the circuit constraint of the dynamic ZKP, verifies whether the hash value of the script output is equal to the output hash value, verifies whether the output identifier is generated based on the verification of the script output, and verifies whether the sum of the input-output hash value and the previous hash value is equal to the current hash value, the input-output hash value being obtained by hashing the sum of the script input and the script output; If all are correct, the AI Agent determines that the execution process of the RPA script is correct. 6.The method of verifying an RPA script of claim 4, wherein, If the execution data further includes an execution timestamp, the AI Agent verifies the correctness of the execution process of the RPA script by using the dynamic ZKP, including: The AI Agent verifies whether the RPA script is executed within a predetermined time range according to the execution timestamp; If yes, the AI Agent determines that the execution process of the RPA script is correct. 7.The method of claim 4, wherein, The AI Agent verifies the consistency of the complete execution flow of all RPA scripts by using all dynamic ZKPs, including: The AI Agent reads the current hash value from the i-th dynamic ZKP and the previous hash value from the i+1-th dynamic ZKP, and verifies whether the current hash value is equal to the previous hash value, i being a positive integer; The AI Agent reads the previous hash value, the input hash value and the output hash value from the i-th dynamic ZKP, and reads the previous hash value from the i+1-th dynamic ZKP, and verifies whether the hash value of the sum of the input hash value and the output hash value is equal to the i+1-th previous hash value after being added to the i-th previous hash value; If all are correct, the AI Agent determines that the execution flow of all RPA scripts is complete. 8.A verification system of an RPA script, characterized by, The verification system includes an AI Agent and an RPA, and the RPA includes an RPA script designer and an RPA script executor; The AI Agent is configured to split the obtained user intent into multiple execution steps; The RPA script designer is configured to generate an RPA script for each execution step; The RPA script executor is configured to generate a static zero-knowledge proof (ZKP) according to version information and script content of the RPA script; The AI Agent is further configured to verify the integrity of the RPA script by using the static ZKP; The RPA script executor is further configured to record execution data during the execution process of each RPA script, and generate a dynamic ZKP according to the execution data; The AI Agent is further configured to verify the correctness of the execution process of the RPA script by using the dynamic ZKP; The AI Agent is further configured to verify the consistency of the complete execution flow of all RPA scripts by using all dynamic ZKPs.

9. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction, which is loaded and executed by a processor to implement the verification method of the RPA script according to any one of claims 1 to 7.

10. A computer device, comprising: The computer device comprises the RPA script verification system of claim 8.

Images