Credential-based universal ai agent for automatically generating and managing credentials

A credential-based AI agent autonomously manages credentials and authentication, addressing the limitations of existing AI agents by securely performing tasks and adapting to risk levels, ensuring both security and convenience in credential management.

WO2026240373A1PCT designated stage Publication Date: 2026-11-19MIN GYENGGWON
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
PCT/KR2026/005516
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2025-05-16
Filing Date
2026-04-21
Publication Date
2026-11-19

AI Technical Summary

Technical Problem

Existing AI agents struggle with securely managing and autonomously performing authentication and credential management across multiple online services, requiring user intervention for identity verification and lacking comprehensive automation capabilities.

Method used

A credential-based general-purpose AI agent that securely stores and manages credentials, automatically performs login and identity authentication, generates and updates credentials, and determines task execution requirements based on predefined thresholds, ensuring both security and convenience.

Benefits of technology

Enables complete automation of credential management and authentication processes without user intervention, balancing security and convenience by dynamically adjusting thresholds and minimizing unnecessary information disclosure.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure KR2026005516_19112026_PF_FP_ABST
    Figure KR2026005516_19112026_PF_FP_ABST
Patent Text Reader

Abstract

The present invention discloses a credential-based universal AI agent capable of automatically generating and managing credentials, which automatically generates and updates a user's credentials and autonomously processes authentication and task performance for various services by using the credentials. The credential-based universal AI agent comprises: a credential storage for storing and protecting one or more credentials for the user; an AI agent task module in charge of authenticating an online service and performing a task; and a credential management unit that controls the AI agent task module to retrieve, from the credential storage, a credential required when the authentication is performed, or to extract the corresponding credential from a physical authentication device, wherein the AI agent task module accesses the online service according to a general-purpose task instruction given by the user, and if login or identity authentication is required before using the online service, automatically performs the login or identity authentication by using the credential secured through the credential management unit, thereby enabling complete automation without user intervention.
Need to check novelty before this filing date? Find Prior Art

Description

Credential-based general-purpose AI agent that automatically generates and manages credentials

[0001] The present invention relates to the field of credential-based artificial intelligence (AI) agents, and more specifically, to a general-purpose AI agent that automatically generates and updates user credentials (e.g., passwords, authentication tokens, etc.) and utilizes them to autonomously handle authentication and task execution for various online services.

[0002] In the modern digital environment, users must create accounts and log in to numerous websites or applications to use online services. However, managing multiple strong passwords is very cumbersome because credentials, such as passwords, must be remembered for each account.

[0003] Due to these difficulties, many users tend to use the same or similar passwords for all services or opt for easy passwords, which poses a significant security risk. If the password for one account is compromised, other accounts using the same password can also be threatened in a chain reaction.

[0004] Therefore, to enhance the convenience of credential management, password managers or single sign-on (SSO) systems have been developed. In this regard, U.S. Patent No. 7155739 discloses a technology that stores a user's various website login credentials on a central server and automatically inputs them when needed to enable one-click login to multiple websites. This prior patent securely stores credentials (ID, password, etc.) registered once by a user and assists in login by automatically or semi-automatically submitting the credentials when the user wishes to use the relevant service.

[0005] Furthermore, with the advancement of ID management and SSO systems in enterprise environments, technologies have emerged where a single central system manages user accounts for various services within the company and automatically changes passwords to strong values ​​periodically. For example, enterprise identity management services enhance security by creating user accounts for each system, setting unique strong passwords, and updating them to random values ​​at fixed intervals to propagate them to all client devices.

[0006] In addition, Robotic Process Automation (RPA) technology, which automates repetitive and routine tasks, is also being utilized in credential management. For example, there are instances where software robots automatically issue temporary credentials and initiate workflows when creating new accounts during the employee onboarding process. According to a report by Bista Solutions, the onboarding process can be streamlined by RPA bots automatically generating credentials, such as initial passwords, and granting permissions while creating new employee accounts. As such, there have been attempts to generate and manage credentials through automation technology.

[0007] Recently, with the growing interest in AI agents, there has been an increasing number of cases where these agents perform online tasks (e.g., information retrieval, transactions, service usage, etc.) on behalf of users. For instance, the latest large-scale language model-based agents have become capable of autonomous behavior, such as collecting information from various websites, interacting with other software APIs, and even accessing shopping websites to make purchases, all in response to user instructions.

[0008] In this context, for AI agents to act on behalf of humans, they must provide authentication methods like humans to log into services or prove their identity, which raises new security and privacy issues. While existing Multi-Factor Authentication (MFA) systems generally require direct user intervention at critical stages (e.g., entering mobile phone OTPs, fingerprint authentication), there is a need to minimize or automate these processes for the smooth operation of autonomous agents.

[0009] Furthermore, many online services require additional identity verification procedures through telecommunication carriers or accredited identity verification authorities (hereinafter referred to as "Identity Providers" (IdPs)) in addition to the login process itself. For example, when using government civil service applications, users must verify their identity through processes such as mobile phone authentication or the submission of an accredited certificate; this is accomplished through the service's integration with a third-party Identity Provider. Since these additional identity verification steps typically require manual input by the user for personal security reasons, they serve as an obstacle to the complete automation of AI agents.

[0010] Furthermore, active research is being conducted on technologies to securely perform authorization delegation to AI agents. A study by Tobin South et al. on arxiv.org, published in 2024, proposes a framework that extends existing authentication protocols using OAuth 2.0 and OpenID Connect to allow human users to delegate a certain range of authority to AI agents and issue agent-specific credentials and metadata. As such, various prior technologies and research exist internationally to address the issues of authentication and authorization management for AI agents.

[0011] As described above, existing methods have generally focused on generating and updating credentials independently of user instructions, such as credential security management or external credentialing, or have concentrated on controlling AI agents that cannot intervene in user authentication procedures. In the future, there is a continuously growing demand for a comprehensive, universal agent system capable of autonomously generating and updating individual user credentials for general user instruction execution, while simultaneously performing all authentications securely and completing instructions without user intervention when external authentication is required.

[0012] The present invention was invented in response to the aforementioned necessity, and its purpose is to provide a general-purpose AI agent that operates based on credentials.

[0013] In particular, the objective of the present invention is to provide a credential-based general-purpose AI agent that is implemented on a user-side terminal device or an external device, securely stores and manages credentials such as passwords, authentication tokens, and digital certificates related to a user, and processes subsequent tasks such as automatically logging in or automatically performing identity authentication by utilizing said credentials without user intervention when performing various tasks according to user instructions.

[0014] In addition, another objective of the present invention is to provide an AI agent capable of automating the automatic generation and updating of user credentials by the agent itself.

[0015] In addition, another objective of the present invention is to ensure both safety and convenience by evaluating the type and risk level of a task to be performed according to a user's instructions and determining, based on a predefined threshold, whether the performance of the task requires user approval.

[0016] The technical problems to be solved by the present invention are not limited to those mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art to which the present invention belongs from the description below.

[0017] The above objective is achieved by a credential-based universal AI agent according to an embodiment of the present invention, comprising: a credential storage unit that stores and protects one or more credentials for a user; a physical authentication device capable of extracting one or more credentials for a user; an AI agent task module responsible for authentication and task execution for the online service; and a credential management unit that controls the AI ​​agent task module to retrieve credentials required for authentication from the credential storage unit or to extract credentials from the physical authentication device; wherein the AI ​​agent task module is configured to access the online service according to universal task instructions requested by the user, and to automatically perform login or identity authentication using credentials obtained through the credential management unit when login or identity authentication is required before using the online service, thereby enabling complete automation without user intervention.

[0018] At this time, the credential management unit may control the login or identity authentication by automatically generating a new credential and adding it to the credential storage if the credential required for the login or identity authentication does not exist in the credential storage, or by terminating the login or identity authentication if the credential is not extracted from the physical authentication device.

[0019] The credential management unit can automatically update credentials stored in the credential storage or extracted from the physical authentication device if they are scheduled to expire or are vulnerable to security breaches.

[0020] The credential storage above may include a security module that encrypts and stores the user's credentials and decrypts and provides the credentials only when there is an authenticated request from the AI ​​agent work module or user approval.

[0021] In addition, the general-purpose AI agent according to an embodiment of the present invention may be implemented on a user-side terminal device or an external device.

[0022] In addition, the general-purpose AI agent according to an embodiment of the present invention may further include a threshold determination unit that evaluates the type and risk level of task execution for a general-purpose task requested by a user through the online service and determines whether the task requires user verification based on a predefined threshold.

[0023] At this time, the threshold determination unit sets a user intervention threshold requiring user verification based on a predefined risk standard according to the attributes of the task, transaction amount, or security level, and determines whether the task requested by the user exceeds the user intervention threshold before executing the task requested by the user through the AI ​​agent task module, and if it exceeds it, requests prior approval or verification from the user and receives a response.

[0024] In addition, the threshold determination unit can dynamically adjust the threshold based on a service policy or real-time judgment, and allows the operation to proceed autonomously without a user confirmation request by exceptionally adjusting the threshold even when the risk level of the operation is high.

[0025] The credential management unit further includes a configuration that restricts the use of credentials not required when activating the AI ​​agent, so that if a specific credential is not required for the task to be performed, access to that credential is intentionally denied or the credential storage is kept locked.

[0026] The above credential includes one of an account and password newly issued through its own algorithm, a one-time authentication code, or a digital signature, or is a hardware security key generated by the physical authentication device including an OTP card, a USB token, or a FIDO security key.

[0027] If the above AI agent work module is pre-registered by directly connecting the physical authentication device during the initial setup phase of the above credential, it can autonomously perform online authentication using the pre-registered hardware security key after registration.

[0028] The above AI agent work module includes a bootstrap algorithm that self-modifies or modifies the system configuration of the general-purpose AI agent through a process of repeating learning, inference, generation, and execution regarding authentication protocols and authentication processes during login or identity authentication between a user terminal and an online service, and can perform recursive self-improvement and automatic authentication through the bootstrap algorithm.

[0029] According to the AI ​​agent of the present invention, when accessing an online service and performing various tasks in accordance with user instructions, it automatically performs login or identity authentication using the user's credentials without user intervention and processes the work. This resolves the issue of credential control, which was a hindrance to full automation in the past, and automatically completes general-purpose tasks required by the user, thereby providing a significant effect of satisfying both credential security and user convenience.

[0030] In addition, according to the AI ​​agent of the present invention, when logging into a web service or application, or online identity authentication, is required on behalf of the user, it has the effect of enabling complete automation without human intervention by automatically generating and updating the user's personal credentials.

[0031] In addition, according to the AI ​​agent of the present invention, by determining whether user approval is required based on a predefined threshold according to the type and risk level of the task, the AI ​​agent automatically performs tasks or routine tasks with a risk level lower than the predefined threshold without asking the user, while requesting prior approval from the user for critical tasks that exceed the threshold, thereby enhancing safety.

[0032] Furthermore, by dynamically adjusting thresholds and making decisions that allow for automatic execution without user approval based on real-time judgment even when thresholds are exceeded, it provides the effect of maintaining a balance between convenience and safety through flexible threshold adjustment management.

[0033] For example, according to the AI ​​agent of the present invention, when a user commands, "Check the bill received in my email and pay the amount," the AI ​​agent logs into the email account (generates or updates credentials if necessary), views the attached bill, and proceeds with payment via a registered card or bank account. Here, the AI ​​agent moves between external online services, such as email services and payment gateway services, and utilizes the necessary credentials for each. For instance, if the payment amount exceeds a threshold set by the user, the system may obtain user approval or have the user perform the necessary authentication immediately before payment; if the system determines the transaction to be secure based on learned patterns, the agent may control the process to complete the payment immediately. Once all tasks are completed, the AI ​​agent reports the processing result to the user and can automatically perform follow-up actions, such as terminating the authentication session used for payment or updating the password.

[0034] FIG. 1 is a configuration diagram of a service system using a credential-based general-purpose AI agent according to an embodiment of the present invention.

[0035] FIG. 2 is an operation diagram showing the interaction between the systems of FIG. 1.

[0036] FIG. 3 is an operational diagram showing the connection relationship between a general-purpose AI agent and a physical authentication device according to an embodiment of the present invention.

[0037] FIG. 4 is a detailed configuration diagram of a general-purpose AI agent according to an embodiment of the present invention.

[0038] FIG. 5 is a flowchart illustrating the process of credential generation and task execution of a general-purpose AI agent according to an embodiment of the present invention.

[0039] Embodiments of the present invention will be described in more detail below.

[0040] The term 'Credential' as used in the present invention is also referred to as a credential and refers to information used to verify the identity of a user of an information system and to determine whether the user is authorized to use it. Cryptographic information used to verify whether a user is an authorized person (e.g., password), a public key / private key pair used in a public key cryptography algorithm, a public key certificate issued by an electronic signature certification business operator, relevant information from a trusted root certification authority (e.g., KISA top-level certification authority), authorization information, etc. may be included.

[0041] The 'Agent' mentioned in this invention refers to a function that understands intent based on the context and executes processing through independent judgment; in other words, it is a program that executes a specific purpose upon locating a target web page, and it is a concept proposed as a next-generation human-computer interface. It is also being proposed as a technology for effectively utilizing various services on the Internet. The objective of this technology is to provide more advanced services by capturing network services using distributed agents. Search robots, which automatically collect information on the Internet and database it, are also a type of agent.

[0042] In the context of the "universal AI agent" mentioned in this invention, the term "universal" refers to the entire range of online operations, such as information retrieval, transactions, and service usage. To perform these "universal" operations across this entire range, user approval must be taken into account for certain tasks depending on the situation. In this invention, the concept of a "threshold" is introduced to establish a point where user approval or confirmation can be involved. Such a "universal" AI agent can collect information from various websites according to user instructions, and an arbitrary agent may also perform tasks such as accessing shopping websites to proceed with purchases.

[0043] The general-purpose AI agent of the present invention is implemented on a user-side terminal device, an external device, or a user-controlled cloud, and is characterized by being able to store, create, and update credentials for a user, and to automatically perform authentication and tasks for various services without user intervention by utilizing the credentials.

[0044] In particular, the general-purpose AI agent of the present invention is designed to automatically manage the credential lifecycle and independently determine the steps requiring user approval based on the type and risk level of the task when performing a task using the credential, thereby ensuring both security and convenience.

[0045] Based on this, the features of the present invention are summarized as follows.

[0046] First, automatic authentication through AI agent task execution secrets

[0047] AI agents can automate the processing of identity verification procedures when requested by online service providers, either directly or through a third-party Identity Provider (IdP).

[0048] In other words, the AI ​​agent can directly submit credentials to the online service provider, or if redirected to a separate IdP page requiring OTP code entry or certificate verification, the AI ​​agent can handle the process on behalf of the user. In special cases, the system is also configured to integrate with legacy authentication modules that, according to the policies of the service provider or third-party Identity Provider (IdP), require credentials to be received directly from the user and submitted to the service provider or IdP. Finally, the system ensures that the agent's service login or transaction is completed upon successful authentication and terminated upon failure.

[0049] Second, automatic generation and updating of credentials associated with AI agent task execution.

[0050] The AI ​​agent according to an embodiment of the present invention enables the complete automation of AI-based task execution by automatically generating and updating user credentials required for online identity authentication on behalf of the user when using an online service.

[0051] At this time, the AI ​​agent according to an embodiment of the present invention monitors the user's credentials for each service and automatically renews them if they are scheduled to expire or are vulnerable to security breaches, and automatically generates credentials if the credentials required for a new service the user intends to use do not exist. For example, the AI ​​agent may periodically check the password strength of each account and change it to a strong, random password if necessary, and perform actions such as creating an account and setting a random initial password if the user is encountering the service for the first time.

[0052] This provides the effect of maintaining strong and unique credentials for all accounts.

[0053] Third, autonomous decision-making based on user intervention thresholds

[0054] The present invention evaluates the types and risk levels of tasks for online services and manages situations requiring user approval using thresholds. Consequently, for tasks with a risk level lower than a predefined threshold or routine tasks, an AI agent automatically performs them without asking the user, while for critical tasks exceeding the threshold, it requests prior approval from the user to enhance safety.

[0055] Furthermore, the AI ​​agent according to an embodiment of the present invention can make the most efficient decisions even in situations where the threshold is exceeded by dynamically adjusting the threshold or making judgments in real time during the operation. For example, in the case of financial transactions involving very large amounts, a user approval procedure is required in principle; however, if the system determines through its own trust model that urgent processing is necessary and safe, it may exceptionally decide to proceed automatically without user approval. Through such flexible threshold management, a balance between convenience and safety can be maintained.

[0056] Fourth, minimum disclosure of necessary information (principle of least authority)

[0057] An AI agent according to an embodiment of the present invention uses credentials only when necessary for performing a task, and is provided so as not to expose or utilize said credentials when unnecessary.

[0058] For example, instead of opening the user's entire credential repository in bulk upon activation, it determines which credentials are required for a specific task and selectively uses only the essential ones. Sensitive credentials that are not used during the process remain encrypted, reducing the risk of potential leakage. This aligns with the Zero Trust principle and controls the AI ​​agent to prevent it from exercising excessive authority.

[0059] Fifth, integration of hardware-based authentication methods

[0060] The present invention integrates physical authentication means, such as a Hardware Security Module (HSM), an OTP generator, and a FIDO2 security key, under the management of an agent. When a user connects and registers a physical authentication device to the AI ​​agent during initial setup, the AI ​​agent can subsequently automatically acquire one-time codes or cryptographic signatures generated through the physical authentication device and use them as credentials during the online authentication process. Through this, the AI ​​agent can software-control and perform even authentication steps that previously required offline or direct user input.

[0061] In particular, AI agents can complete this authentication process without physical human intervention by pre-learning the operating principles of the hardware device or directly controlling the device through a trained model. This enables the complete automation of the authentication process by having the AI ​​agent perform the physical authentication steps that previously required direct user operation.

[0062] Sixth, system change through recursive self-improvement

[0063] A prominent feature of the present invention is that the general-purpose AI agent operation includes the operation of changing each configuration of the general-purpose AI agent.

[0064] The AI ​​agent of the present invention aims to complete specific instructions without user intervention, and user authentication is required to perform specific tasks during the completion process. When the current AI agent is unable to perform these tasks, the invention includes an algorithm that performs a series of bootstrap methods to autonomously modify or adapt the system configuration of the AI ​​agent by repeatedly learning, inferring, generating, and executing authentication metadata, such as all existing authentication protocols and authentication processes. Through the bootstrap algorithm, recursive self-improvement and automatic authentication can be performed.

[0065] This consists of a series of cycles including planning system changes for LLM-based self-improvement, code generation and self-modification, iterative execution loops, reinforcement learning and simulation optimization, and tool and API integration. By enabling system changes, including self-improvement through this cycle, to be carried out without user intervention, it is designed to perform the role of a fully universal AI agent.

[0066]

[0067] The features of the present invention will be examined in detail below with reference to the attached drawings.

[0068] FIG. 1 is a configuration diagram of a service system using a credential-based general-purpose AI agent according to an embodiment of the present invention, and FIG. 2 is an operation description diagram showing the interaction between the systems of FIG. 1.

[0069] A credential-based general-purpose AI agent (200) (hereinafter referred to as 'AI agent') according to an embodiment of the present invention plays a role in reliably performing various tasks requiring user authentication while minimizing user intervention, between a user terminal (100) that requests and receives an online service desired by the user and service servers (300) that provide various online services.

[0070] These AI agents may be implemented within the user terminal (100) or installed on an external device and connected to the user terminal (100) via the external device. Alternatively, they may be implemented in a user-controlled cloud.

[0071] In this embodiment, when a user device (100) sends a work command or request to an AI agent (200) (①), the AI ​​agent (200) connects to the corresponding online service server (300) on behalf of the user device (100) and performs the work instructed by the user device (100). At this time, if the online service server (300) requires user login or identity authentication before using the service, it requests user approval from the AI ​​agent (200) (②), and the AI ​​agent (200) internally accesses the credential storage (240) to search for the necessary credential (or credentials) or, if not stored in the credential storage (240), creates a new credential (③). The AI ​​agent (200) submits the credentials obtained in this way to the service server (300) on behalf of the user terminal (100) (②), and when authentication is completed by the service server (300), it provides final approval to the AI ​​agent (200) (⑥), thereby performing automatic login or automatic identity authentication and processing the task instructed by the user terminal (100).

[0072] Alternatively, if the service server (300) requires authentication of the identity claimed by the user through a separate third-party identity provider (IdP) (400), the AI ​​agent (200) communicates with the IdP (400) instead of the service server (300) to submit the user's credentials required for login or identity authentication to the IdP (400) (5-1), and when authentication is completed at the IdP (400), the authentication result is transmitted directly or indirectly to the service server (300) (5-2). Through the above procedure, the service server (300) can finally approve or reject the service credentials of the AI ​​agent (200) based on the received authentication result (⑥). Accordingly, the AI ​​agent (200) receives the approval or rejection result from the service server (300) and can perform authentication and work processing accordingly.

[0073] As can be seen in FIG. 1, the AI ​​agent (200) acts as an agent that receives instructions from the user terminal (100) and takes action on the service server (300), and in this process, it manages the credential storage (240) and operates as a hub that communicates with the user terminal (100).

[0074] The credentials in the present invention may include an account and password newly issued through an algorithm embedded in the AI ​​agent, or one of a one-time authentication code or a digital signature. Alternatively, they may include a security key using a hardware authentication device (or authentication means), such as an OTP (One Time Password) card, a USB token, or a FIDO (Fast Identity Online) key. In the latter case, if a hardware authentication device, such as an OTP card or a FIDO key, is pre-connected to the AI ​​agent and pre-registered during the initial configuration phase of the credentials, the AI ​​agent obtains a one-time authentication code or generates a digital signature from the hardware authentication device whenever necessary after registration and utilizes it in the authentication process of an external service. At this time, the AI ​​agent can obtain the necessary code by communicating with the device software-wise without additional user operation.

[0075] Meanwhile, the AI ​​agent (200) can also be linked with the hardware authentication device (or physical authentication device) when applying a security key using a hardware authentication device, such as an OTP card, USB token, or FIDO (fast Identity Online) security key, as a credential.

[0076] Figure 3 is a diagram showing the connection relationship between an AI agent and a physical authentication device.

[0077] Referring to FIG. 3, when a service server (300) or an idP (400 in FIG. 2) requires additional authentication, such as a one-time authentication code (OTP) or a digital signature (Signature), an AI agent (200) produces credentials, such as obtaining the one-time authentication code or generating a digital signature of the user, through a registered physical authentication device (500), and submits the produced credentials, such as the one-time authentication code or digital signature, to the service server (300) or the IdP (400 in FIG. 2). Upon receiving this, the service server (300) or the IdP (400 in FIG. 2) performs authentication through an authentication protocol. Based on the authentication result, the AI ​​agent (200) receives a notification from the service server (300) regarding whether the task is approved, thereby allowing the authentication step through the physical authentication device (500) to be automated without user intervention.

[0078] FIG. 4 is a detailed configuration diagram of an AI agent according to an embodiment of the present invention.

[0079] The AI ​​agent (200) according to an embodiment of the present invention is an agent that independently determines a user's request (instruction), accesses various online service servers sequentially or simultaneously, and comprehensively completes the task requested by the user.

[0080] To this end, the AI ​​agent (200) includes a communication module (not shown), an AI agent work module (210), a credential management unit (220), a threshold determination unit (230), and a credential storage unit (240).

[0081] The communication module performs data communication with a user terminal (100 in FIG. 2), an external service server (300 in FIG. 2), as well as a third-party identity verification provider (IdP, 400 in FIG. 2) and a physical authentication device (500 in FIG. 3) through a defined communication protocol.

[0082] The credential store (240) stores and protects one or more credentials for a user.

[0083] At this time, the credential storage (240) may encrypt and store the user's credentials. In this case, the credential storage (240) may further include a security module that encrypts the user's credentials and decrypts and provides the credentials when there is an authenticated request from the AI ​​agent work module (210) or user input.

[0084] The credential management unit (220) controls the AI ​​agent work module (210) to retrieve the credential required for performing the work from the credential storage (240) or to extract the credential from the physical authentication device.

[0085] The credential management unit (220) can control the automatic generation of new credentials and addition to the credential storage (240) if the credentials required for login or identity authentication do not exist in the credential storage (240).

[0086] Additionally, the credential management unit (220) can automatically update the credential if the credential stored in the credential storage (240) or extracted from the physical authentication device is scheduled to expire or is vulnerable to security breaches.

[0087] Additionally, the credential management unit (220) can create a new account by performing the membership registration process for the service on behalf of the user, set a randomly generated strong password, and store this information in the credential storage (240) even if there are no credentials for the account to be logged into the credential storage (240).

[0088] Alternatively, the credential management unit (220) may control the termination of login or identity authentication if credentials are not extracted from a pre-registered physical authentication device.

[0089] Additionally, the credential management unit (220) may further include a configuration that restricts the use of credentials that are not required when activating the AI ​​agent (200), thereby intentionally denying access to the credentials or keeping the credential storage (240) locked when specific credentials are not required for the task to be performed.

[0090] The AI ​​agent work module (210) is responsible for both authentication and task execution for various online services.

[0091] The AI ​​agent work module (210) accesses the corresponding online service according to a general-purpose work instruction requested by a user terminal (100 in FIG. 2) and automatically performs login or identity authentication before using the service through credentials obtained via the credential management unit (220), thereby enabling full automation without user intervention.

[0092] To this end, the AI ​​agent work module (210) submits the user's credentials obtained by the credential management unit (220) to the service server of the relevant online service or the relevant third-party identity verification provider (IdP), and receives a verification result for identity authentication from the service server. For example, the AI ​​agent work module (210) may receive a verification result for identity authentication by directly submitting the credentials required for login or identity authentication to the service server, or the AI ​​agent work module (210) may receive the credentials to the relevant IdP, and receive the authentication result from the service server when the authentication result from the IdP is transmitted directly or indirectly to the service server. Accordingly, the AI ​​agent work module (210) receives an approval or rejection result from the service server (300) and performs authentication and work processing accordingly. This procedure is illustrated in detail in FIG. 2.

[0093] The threshold determination unit (230) evaluates the type and risk level of a task requested by a user through an online service and determines whether the task requires user confirmation based on a predefined threshold.

[0094] Additionally, the AI ​​agent work module (210) can perform authentication by linking with a legacy authentication module that must receive credentials directly from the user and submit them to a service server or IdP. For example, if a service provider or a third-party identity authentication provider does not allow the AI ​​agent and applies an authentication policy that compels human action, automatic authentication may not be performed by the AI ​​agent. Even if the existing authentication procedure is executed inevitably due to such a policy, the AI ​​agent work module (210) can control the process to enable automatic authentication by linking with a legacy authentication module (e.g., a fingerprint recognition module, etc.) and intervening in the pre- and post-authentication processes.

[0095] Specifically, the threshold determination unit (230) sets a user intervention threshold (hereinafter, threshold) requiring user verification based on a predefined risk standard according to the attributes of the task, transaction amount, or security level, and determines whether the task exceeds the threshold before executing the task requested by the user through the AI ​​agent task module (210). In the case of a critical task that exceeds the threshold, the user is requested to pre-approval or verification to enhance safety, and in the case of a low-risk task or routine task that does not exceed the threshold, the task is automatically executed without requesting any verification from the user.

[0096] Here, the method of requesting prior approval or confirmation from the user may consist of a procedure of notifying the user of a message regarding the nature of the task and obtaining confirmation from the user. Alternatively, a secondary additional identity verification procedure may be required.

[0097] Additionally, the threshold determination unit (230) can dynamically adjust and apply the user intervention threshold based on information collected during the operation. In this case, the threshold determination unit (230) can autonomously proceed with the operation without a request for confirmation (or approval) from the user based on a pre-set policy or real-time judgment, even if the risk level of the operation is high.

[0098] For example, in the case of operations involving a payment process, such as very large financial transactions or the purchase of reserved airline tickets, a user approval procedure is required in principle; however, if the system applies a trust model and determines that urgent processing is necessary and simultaneously safe, it may exceptionally decide to proceed automatically without user approval. Through such flexible threshold management, both convenience and safety can be secured. Usually, securing safety leads to a decrease in convenience, and emphasizing convenience leads to a decrease in safety (security); however, the present invention has the advantage of securing both convenience and safety.

[0099] Meanwhile, when the AI ​​agent work module (210) receives a request from the service server to authenticate a user's identity through a third-party Identity Provider (IdP), it submits credentials related to the user's identity to the IdP. When the service server, having received the authentication result directly or indirectly from the IdP, decides to approve the service credentials, the AI ​​agent work module (210) can automate the execution of post-credential work online.

[0100] FIG. 5 is a flowchart illustrating the process of generating credentials and performing tasks for a general-purpose AI agent according to an embodiment of the present invention.

[0101] First, when the user terminal launches the AI ​​agent app or accesses the corresponding webpage and the user inputs the desired service task, the AI ​​agent is activated and starts operation (S1).

[0102] Next, the AI ​​agent determines whether login or identity authentication is required for the task to be performed by input from the user terminal (S2).

[0103] Next, if authentication is required, check what credentials are required (S3).

[0104] In this S3 step, if authentication is unnecessary, the task execution step (S8) is performed immediately without accessing the AI ​​agent's credential store and physical authentication device. At this time, the AI ​​agent automatically proceeds with and completes the requested task without using the user's credentials at all.

[0105] Conversely, in the S3 stage, if authentication is required, access the credential store or physical authentication device to utilize the user's credentials and prepare to use the necessary credentials.

[0106] Next, the AI ​​agent checks whether the credentials required for authentication exist in the storage or are extracted by linking with a physical authentication device (S4).

[0107] If, upon verification at step S5, the credential does not exist in the credential repository, the AI ​​agent proceeds to step S10, which generates a new credential. This process involves issuing a new service account or generating a password, and although it is shown as a single block in Fig. 5, it may actually include detailed procedures such as account registration and receiving a temporary password. For example, the AI ​​agent may perform the membership registration process for the service on behalf of the user to create a new account, set a strong password generated randomly, and then store this information in the credential repository. Through this, the necessary credentials can be immediately obtained and utilized in subsequent procedures.

[0108] Upon verification, if the credentials are not extracted from the physical authentication device at step S5, control is implemented to terminate the operation with the corresponding server.

[0109] As a result of the verification at step S5, even if the credential exists in the credential storage or can be extracted from the physical storage device, the AI ​​agent can check the expiration time of the credential or additionally check whether the security is vulnerable (S6).

[0110] If, through this validation, an update is required based on certain criteria (e.g., the expiration date is approaching or there is a security vulnerability), the corresponding credential is updated or regenerated (S20).

[0111] Conversely, if the validation check confirms that the credentials are still valid and working properly, the process proceeds to the next step without the renewal procedure.

[0112] Since the above credential validation step (S6) and renewal step (S20) may be performed solely for the purpose of renewing the credential, in addition to not being performed according to the above series of authentication procedures, they may be performed at every renewal cycle or before and after other steps, without being limited to the order of the process.

[0113] Next, the AI ​​agent checks the user intervention threshold, which takes into account the sensitivity, risk, and importance of the task, to determine whether user approval is required for the task (S7).

[0114] For example, in the case of large transactions or high-risk operations, additional verification may be required as the threshold is exceeded. This process is designed to enhance safety by determining whether the user intervention threshold has been exceeded before the AI ​​agent logs into an external service and performs the actual task execution. This process is not limited to the S7 step sequence and can proceed regardless of order, such as during the credential search and extraction process, as long as the task execution is performed beforehand.

[0115] In step S7, if the threshold is exceeded, a user approval request step (S30) may be performed. This can be performed by notifying the user and obtaining confirmation.

[0116] The user approval request step (S30) in this embodiment can be performed in two ways depending on the system settings.

[0117] As a static-based approach, the AI ​​agent immediately suspends task execution and asks the user for approval to proceed with tasks that exceed a predefined threshold. If approval is confirmed by the user, it proceeds to the next step; if denied, the task is stopped.

[0118] As a dynamic autonomous approach, the AI ​​agent independently determines whether to proceed with the current task by dynamically adjusting threshold criteria based on pre-defined policies or AI algorithms. For example, if the AI ​​algorithm is configured to allow autonomous decision-making by the AI ​​agent when a threshold is exceeded, the agent may evaluate the risk level of the current situation and automatically execute a task without user approval—exceptionally, even if it would normally require it—or conversely, decide to request user approval for a task that could normally be performed without it. This involves making a real-time decision on whether to proceed or cancel before the task is executed; if the decision is to proceed, the next step is executed, and if the decision is to cancel, the task is terminated.

[0119] In step S7, if the threshold is not exceeded, the AI ​​agent automatically continues the task without requesting confirmation from the user.

[0120] That is, the AI ​​agent attempts to log in by connecting to the target external service server with the secured credentials, and if authentication is successful, immediately performs the desired task on the service (S8).

[0121] At this time, if the service server requires a third-party identity verification procedure (e.g., OTP code input or biometric authentication), the AI ​​agent processes the identity verification procedure in parallel. (Proceeds identically from S4 to S6)

[0122] The third-party identity verification process can be conducted through third-party Identity Providers (IdPs), such as telecommunications carriers, financial institutions, credit card companies, and government agencies. For instance, this can involve mobile phone identity verification via telecommunications carriers, accredited certificate verification via financial institutions, card identity verification via credit card companies, bank account identity verification via banks, or Visa identity verification via government agencies. When an IdP requests credentials, the AI ​​agent retrieves them via a credential repository or physical authentication device, submits them to the IdP, and completes the identity verification. Based on the IdP's verification results, the service server determines whether to allow the operation to proceed; if authentication is successful, the AI ​​agent continues with the original task, while if it fails, it stops the operation.

[0123] We examine preferred embodiments of this process.

[0124] As an example, a case in which a personal assistant-type AI agent is implemented on a user's smartphone is described.

[0125] This agent app pre-stores account information frequently used by the user, such as email, social media, and internet banking, and protects it within the device's Secure Enclave using strong encryption like AES-256. When a user issues a voice command or text instruction such as, "Book a flight for next week's schedule and send a confirmation email to my company email," the agent accesses the flight booking site and searches for tickets matching the user's profile. It automatically retrieves or generates the necessary credentials (airline website login information, payment card details, etc.). For instance, if the user visits the airline site for the first time, the agent automatically performs the registration process to create an account and generates and sets a unique password. After selecting a flight, the agent retrieves the company expense card information to proceed with the payment during the payment stage; if the payment amount is large, it sends a confirmation message to the user via smartphone notification asking, "Would you like to proceed with the flight payment?" based on a threshold set in the app. If the user presses the confirmation button, the agent immediately completes the payment and automatically sends a confirmation email using the company email account's SMTP. In this scenario, the agent navigated across three external service servers (airline website, payment gateway, and email), utilizing the necessary credentials (newly created airline account, saved payment card, and saved email login information) for each in a timely manner. Furthermore, it maintained a balance between security and convenience by obtaining user approval at critical transaction stages based on threshold settings. Upon completion of the task, the agent securely stored the generated airline account information and recorded the details of the actions performed for future use in a log.

[0126] As another embodiment, a case in which a personal assistant-type AI agent is implemented in the form of an app on a user's smartphone is described.

[0127] The AI ​​Agent app pre-stores account information frequently used by the user, such as email, social media, and internet banking, and protects it within the device's Secure Enclave (hardware security technology built into the smartphone) using strong encryption like AES-256. When a user instructs via voice command or text, "Book a flight according to next week's schedule and send a confirmation email to my company email," the AI ​​Agent app accesses the flight booking site and searches for flights matching the user's profile. At this stage, the AI ​​Agent app automatically retrieves or generates the credentials (airline website login information, payment card details, etc.) required to log in to the booking site and purchase the flight. For example, if the user visits the airline site for the first time, the AI ​​Agent app automatically performs the registration process to create an account and generates and sets a unique password. Subsequently, the user selects a flight and proceeds with payment by retrieving company expense card information during the payment stage. If a threshold is set for large payment amounts requiring the AI ​​Agent app to request payment approval, it sends a confirmation via smartphone notification asking, "Would you like to proceed with the flight payment?" When the user presses the confirmation button, the agent app immediately completes the payment and automatically sends a confirmation email using the company email account's SMTP.

[0128] In this scenario, the AI ​​agent app navigates across three external service servers (airline website, payment gateway, and email), utilizing the necessary credentials (newly created account, saved card, and saved email login) for each in a timely manner. Furthermore, it maintains a balance between security and convenience by obtaining user approval based on thresholds. Upon completion of the task, the agent app securely stores the generated airline account information and can also log usage history, such as email details, for future use.

[0129] As another embodiment, an example of applying the present invention to an enterprise Robotic Process Automation (RPA) agent is described.

[0130] For example, the IT management department of a large corporation can configure an AI agent to collectively manage various work accounts (email, messenger, ERP, etc.) for employees. When a new employee's information is registered in the internal HR system, the AI ​​agent detects this and immediately creates accounts for internal systems such as email and messenger, automatically issuing and setting initial passwords. These passwords are changed periodically according to established rules, and the agent also performs the task of collectively deleting or deactivating the relevant accounts when an employee leaves the company. In particular, for tasks requiring a multi-stage approval process (e.g., approval of internal approval documents), this AI agent requests approval from executives according to a predefined workflow and processes subsequent tasks once approved. If an executive is absent, the system may utilize dynamic decision-making functions, such as automatically processing approval after a waiting period based on established rules or delegating the request to a proxy approver. In such an enterprise environment, the AI ​​agent of the present invention combines credential lifecycle management with business automation, thereby reducing security risks and increasing work efficiency.

[0131] In the following, another embodiment, a method for automating identity verification procedures for government or financial services using the AI ​​agent of the present invention is described.

[0132] For example, let's assume a user commands an AI agent, "Log in to the government civil service portal and obtain my Resident Registration Certificate." Generally, using such government services requires undergoing a separate identity verification process (e.g., mobile phone authentication, accredited certificate authentication) to confirm the user's identity. In this case, the AI ​​agent can perform identity verification by inputting and submitting the information required during the identity verification stage of the portal site on behalf of the user. Specifically, the AI ​​agent enters basic information such as the username and Resident Registration Number on the portal and selects the mobile phone authentication method. When the portal connects to the mobile carrier's authentication page and sends an SMS authentication code to the user's mobile phone, the AI ​​agent detects the SMS on the smartphone or receives the code from the user and submits it to the authentication page. Subsequently, upon receiving a "Identity Verification Complete" response from the mobile carrier, it proceeds with the portal's login process. If the user is required to use an accredited certificate, the AI ​​agent can perform authentication by utilizing a digital certificate that is either stored in advance or connected by the user. For instance, if the certificate is stored on a USB security token connected to the user's computer, the AI ​​agent calls that device to generate and submit an electronic signature, thereby completing the identity verification. If the identity verification step through an IdP (e.g., a telecommunications carrier or an accredited certification authority) is successful, the AI ​​agent logs normally into the civil service portal and continues with the procedure for issuing a Resident Registration Certificate; if authentication fails, the operation is stopped, and the user is notified of the cause of the failure. In this way, the AI ​​agent of the present invention can automate or at least simplify the identity verification process, which previously required user intervention, by linking with a third-party identity verification service (IdP).

[0133] The above embodiments merely describe specific applications of the present invention, and the scope of the present invention is not limited thereto. For example, the types of credentials managed by the agent may be extended beyond passwords to include biometric information, security keys, blockchain wallet keys, etc., and threshold-based policies can also be customized for each user. For example, in the case of hardware security keys, a procedure is required during the initial setup phase for the user to directly connect or register the key with the agent; however, once registered, the agent can autonomously perform online authentication using the hardware. Furthermore, the agent of the present invention may operate within a single device or be implemented in a distributed manner as an agent service on the cloud and a lightweight client on a local device. All such changes or variations should be understood to be included within the spirit and scope of the present invention.

[0134] The present invention relates to an AI agent capable of implementing fully automated processing by utilizing user credentials to automatically log in or perform identity authentication without user intervention, and by automatically generating and updating user credentials on its own when necessary. It is applicable to all information and communication-based industrial fields requiring authentication and access control.

[0135] Specifically, the present invention can be utilized to automate the creation, authentication, authorization, and access control of user accounts in corporate account management systems, cloud services, e-commerce platforms, financial service systems, healthcare information systems, etc. In particular, operational efficiency can be improved by automatically performing repetitive authentication procedures required during new user onboarding, inter-system authentication, and service usage.

Claims

1. In an AI agent that accesses an online service and performs a task requested by a user on behalf of said user, A credential store that stores and protects one or more credentials for a user; A physical authentication device capable of extracting one or more credentials for a user; AI agent task module responsible for authentication and task execution for online services; The above AI agent work module includes a credential management unit that controls the AI ​​agent work module to retrieve credentials required for authentication from the credential repository or to extract credentials from the physical authentication device; The above AI agent work module is, A credential-based general-purpose AI agent characterized by being configured to enable complete automation without user intervention by accessing the relevant online service according to general-purpose work instructions requested by the above-mentioned user, and automatically performing login or identity authentication using credentials obtained through the credential management unit when login or identity authentication is required before using the above-mentioned online service.

2. In Paragraph 1, The above credential management department is, A credential-based general-purpose AI agent characterized by controlling to automatically generate new credentials and add them to the credential repository when the credentials required for the login or identity authentication do not exist in the credential repository, or to terminate the login or identity authentication when the credentials are not extracted from a physical authentication device.

3. In Paragraph 1 or 2, The above credential management department is, A credential-based general-purpose AI agent characterized by automatically renewing credentials stored in the credential repository or extracted from the physical authentication device when they are scheduled to expire or are vulnerable to security breaches.

4. In Paragraph 1, The above credential store is, A credential-based general-purpose AI agent characterized by including a security module that encrypts and stores the credentials of the user and decrypts and provides the credentials only when there is an authenticated request from the AI ​​agent work module or user approval.

5. In Paragraph 1, The above general-purpose AI agent is a credential-based general-purpose AI agent characterized by being implemented on a user-side terminal device or an external device.

6. In Paragraph 1, A threshold determination unit that evaluates the type and risk level of task execution for a general-purpose task requested by a user through the above online service, and determines whether the task requires user verification based on a predefined threshold; A credential-based general-purpose AI agent characterized by further including 7. In Paragraph 6, The above threshold determination unit is, A credential-based general-purpose AI agent characterized by setting a user intervention threshold requiring user verification based on a predefined risk criterion according to the attributes, transaction amount, or security class of the above-mentioned task, determining whether the task requested by the user exceeds the user intervention threshold before executing the task through the AI ​​agent task module, and, if it exceeds the threshold, requesting prior approval or verification from the user and receiving a response.

8. In Paragraph 6, The above threshold determination unit is, The above threshold can be dynamically adjusted by service policy or real-time judgment, and A credential-based general-purpose AI agent characterized by enabling the operation to proceed autonomously without a user confirmation request by adjusting the threshold, even when the risk level of the above operation is high.

9. In Paragraph 1, The above credential management department is, A credential-based general-purpose AI agent characterized by additionally including a configuration that restricts the use of credentials not required when activating the AI ​​agent, thereby intentionally denying access to said credentials or keeping the credential repository locked when specific credentials are not required for the task to be performed.

10. In Paragraph 1, The above credentials are, A credential-based universal AI agent characterized by including one of an account and password newly issued through a proprietary algorithm, a one-time authentication code, or a digital signature, or being a hardware security key generated by the physical authentication device including an OTP card, a USB token, or a FIDO security key.

11. In Paragraph 1, The above AI agent work module is, A credential-based general-purpose AI agent characterized by autonomously performing online authentication using a pre-registered hardware security key after registration, when the physical authentication device is directly connected and pre-registered during the initial setup phase of the credential.

12. In Paragraph 1, The above AI agent work module is, A credential-based general-purpose AI agent characterized by including a bootstrap algorithm that self-modifies or modifies the system configuration of the general-purpose AI agent through a process of repeating learning, inference, generation, and execution of authentication protocols and authentication processes during login or identity authentication between a user terminal and an online service, and performing recursive self-improvement and automatic authentication through the bootstrap algorithm.