An artificial intelligence dispatching and running system and method based on control right externalization and one-way interaction constraint

Abstract

This invention discloses an artificial intelligence scheduling and operation system and method based on external control and unidirectional interaction constraints. The system combines a data interaction channel with a security management terminal to enable task execution and data interaction of AI computing units in a controlled environment, ensuring full-process controllability, auditability, and traceability. The system includes a data access unit, isolated AI computing units, a data interaction channel, and a security management terminal. Data is transmitted from the data access unit to the isolated AI computing units through the data interaction channel, which is either a physically unidirectional structure or a logically unidirectional transmission structure constructed by the security management terminal based on access control policies. The security management terminal, as the sole control hub of the system, provides unified control over task scheduling, data interaction, result acquisition, and anomaly handling, ensuring that the isolated AI computing units operate only as passive execution units, lacking task initiation and communication decision-making capabilities. This invention is applicable to environments without or with limited networks, achieving secure, controllable, and auditable operation of the AI ​​system through external control and unidirectional interaction constraints.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence safety scheduling technology, and in particular to a scheduling and operation system and method for achieving safe operation of an artificial intelligence system through external control and one-way interactive constraints. Background Technology

[0002] As artificial intelligence (AI) technology continues to iterate at an increasingly rapid pace, the challenges and pressures faced by relying solely on software-level security mechanisms are constantly growing. However, AI systems cannot overcome the fundamental constraints of physical laws, which dictates that building a secure operating mechanism based on physical isolation is the technical path to ensure fundamental system security. At the same time, many enterprises still have actual business needs for networked collaboration, and a completely disconnected physical isolation solution is insufficient to meet business continuity requirements. Therefore, there is an urgent need for an AI scheduling and operation mechanism that possesses both physical isolation security and supports controlled access to limited networks. Summary of the Invention

[0003] This invention externalizes system control to a security management terminal, structurally depriving the AI ​​computing unit of autonomous task initiation and communication decision-making capabilities, allowing it to exist only as a passive executor, thus constructing a controlled, one-way interactive operating system. One-way or controlled data input transmission is achieved through a data interaction channel, with the security management terminal uniformly controlling task scheduling, data interaction, result acquisition, and anomaly handling, forming a closed-loop control of data input → isolation processing → result output. This invention supports two implementation methods: • In a network-free environment, the AI ​​computing unit is isolated and deprived of general network communication capabilities; • In a limited network environment, only controlled communication within the local area network is allowed, and internet access is prohibited. Beneficial effects

[0004] 1. External control reduces the risk of system loss of control; 2. One-way or controlled interaction ensures data security; 3. Supports both offline and wired network scenarios, balancing security and business needs; 4. Full-process auditing and traceability; 5. Improve the overall security and controllability of the system. Attached Figure Description

[0005] Figure 1 This is a schematic diagram of the system architecture; Figure 2 This is a schematic diagram of the scheduling and operation process. Detailed Implementation

[0006] In this invention, the security management terminal receives task requests from external business systems. The data access unit preprocesses the input data and transmits it to the isolated artificial intelligence computing unit via a data interaction channel. The isolated artificial intelligence computing unit executes artificial intelligence model calculations when trigger conditions are met and returns the results to the security management terminal. Throughout the process, the artificial intelligence computing unit lacks autonomous task initiation and communication decision-making capabilities. In a network-free implementation, the computing unit runs an operating system environment that deprives it of network capabilities; in a limited network implementation, only controlled access via the local area network is allowed. The security management terminal records the entire task execution process, enabling auditability and traceability.

Claims

1. An artificial intelligence scheduling and operation system, characterized in that, include: • Data access unit, used to collect or receive task input data; • An isolated AI computing unit for executing AI model calculations; • A data interaction channel for unidirectional or controlled data transmission from the data access unit to the isolated AI computing unit; • A security management terminal for unified control of task scheduling, data interaction, result acquisition, and end-to-end auditing; The data interaction channel is a physically unidirectional structure, or a logically unidirectional transmission structure built by the security management terminal based on access control policies; the security management terminal is the system's sole control hub; the isolated AI computing unit operates as follows throughout the entire process: (1) It lacks the ability to initiate tasks; (2) Lacks communication decision-making ability; (3) All task execution is triggered or controlled by the security management terminal; (4) All data interactions are completed through the data interaction channel; the system forms a closed-loop control of data input → isolated calculation → result feedback, and all interactive behaviors are controlled, recordable and traceable.

2. The system according to claim 1, characterized in that: The data interaction channel includes data diodes, optical isolation, electrical isolation, physical transport media, or a logical unidirectional channel constructed based on network strategies.

3. The system according to claim 1, characterized in that: The isolated artificial intelligence computing unit triggers task execution by detecting preset data locations or receiving control commands.

4. The system according to claim 1, characterized in that: When a task execution is abnormal or exceeds preset conditions, the security management terminal triggers a processing strategy, including task retry, resetting the computing unit, or manual intervention.

5. The system according to claim 1, characterized in that: The operating system kernel that is isolated from the artificial intelligence computing unit and strips away the general network communication protocol stack is used to prevent the running process from actively establishing external network connections (in the absence of a network).

6. The system according to claim 1, characterized in that: The isolated AI computing unit operates in a controlled network environment, allowing only local area network data interaction with authorized nodes and prohibiting access to the Internet (in a limited network scenario).

7. The system according to claim 1, characterized in that: The data access unit preprocesses the input data, including format conversion, desensitization, or encryption.

8. An artificial intelligence scheduling and operation method, characterized in that, Applied to the system according to any one of claims 1 to 7, comprising the following steps:

1. Receive task requests through the security management terminal; 2. Collect or receive task input data through the data access unit and perform preprocessing; 3. Data is transmitted to the isolated artificial intelligence computing unit via a data interaction channel; 4. Under triggered conditions, the isolated artificial intelligence computing unit performs artificial intelligence model calculations; 5. Obtain and output the calculation results through the security management terminal; 6. Record and audit the entire process of task execution and data interaction; the isolated artificial intelligence computing unit does not have the ability to initiate tasks or make communication decisions autonomously throughout the process.

Images