A dynamic key protection method and system based on AI context perception

Abstract

The present invention relates to the field of key data processing, and in particular to a dynamic key protection method and system based on AI context perception. The method first responds to the user inputting login credentials on the platform side, then calls a preset AI scene recognition module to determine the legitimacy of the platform side page. If the platform side page is legitimate, a short-term token is generated and returned to the platform front end. During the execution of the above process, the method can also identify sensitive call behaviors in real time and call a preset key interception module to intercept sensitive call behaviors. Compared with the existing technology, the method of the present invention combines AI technology to realize credential encryption and abnormal operation interception, solving the problem of low security of existing password management technology.

Description

Technical Field

[0001] The present invention relates to the field of key data processing. More specifically, the present invention relates to a dynamic key protection method and system based on AI context perception. Background Art

[0002] With the development of information technology, network information security has gradually become a concern. When enterprise employees use private / public accounts and keys to log in to network platforms or systems, their password management solutions usually rely on plug-ins or centralized vaults. Among them, Chinese patent application publication number CN119249402A discloses a browser plug-in-based account management method, system, electronic device, and medium. This method relies on plug-ins and AI tools to achieve secure management of user passwords. This password management solution has the following drawbacks:

[0003] It is impossible to intercept sensitive call behaviors (such as screenshots or malicious kernel information / virtual machine theft), and users can often directly obtain plaintext credentials, which may lead to the risk of memory leakage and certain security risks.

[0004] Therefore, the password management in the prior art has the problem of low security. Summary of the Invention

[0005] In order to solve the above-mentioned technical problem of low password management security, the present invention discloses a dynamic key protection method and system based on AI context perception.

[0006] In a first aspect, the present invention discloses a dynamic key protection method based on AI context awareness, comprising:

[0007] S10: In response to the user entering login credentials on the platform, calling a preset AI scene recognition module to determine the legitimacy of the platform page;

[0008] S20: If the platform page is valid, generate a short-term token and return it to the platform front end;

[0009] In any of steps S10 to S20, the method performs the following steps in parallel:

[0010] Identify sensitive call behaviors and call the preset key interception module to intercept sensitive call behaviors.

[0011] Beneficial effect: After the user enters the login credentials on the platform side, if the platform side page is legal, a short-term token is generated and returned to the platform front end to complete the login. The short-term token is not the plaintext key / credential in the prior art, and can effectively prevent plaintext leakage, debugging and screenshot attacks. On this basis, the method of the present invention can also identify sensitive call behaviors in the process of generating short-term tokens, and perform bottom-level interception on sensitive call behaviors to avoid hacker attacks. Compared with the prior art, the method of the present invention realizes credential encryption and abnormal operation interception, solving the problem of low security of existing password management technology.

[0012] Preferably, identifying sensitive call behaviors and calling a preset key interception module to intercept the sensitive call behaviors includes:

[0013] Identify whether the CPU has a tendency to execute sensitive instructions, cryptographic library function call instructions, or abnormal context behavior;

[0014] If so, the preset key interception module is called to intercept the CPU from executing sensitive instructions, cryptography library function call instructions, or execution processes with abnormal context behavior tendencies.

[0015] Furthermore, abnormal contextual behavior tendencies are identified using an AI scene recognition module.

[0016] Beneficial effects: For the interception of sensitive call behaviors, the method of the present invention mainly intercepts three directions: CPU execution of sensitive instructions, cryptographic library function call instructions and abnormal context behavior tendencies. Among them, context behavior is identified in combination with the AI scene recognition module, realizing the effective combination of traditional technology and AI technology, which can more accurately intercept sensitive call behaviors and improve the security of password management technology.

[0017] Preferably, calling a preset AI scene recognition module to determine the legitimacy of the platform-side page includes:

[0018] Obtain contextual behavior features;

[0019] Encode and fuse the context behavior features to obtain the fusion result;

[0020] The fusion results are input into the pre-trained AI scene recognition module to output the safety assessment score;

[0021] If the security assessment score is higher than the security threshold, the platform page is judged to be legal;

[0022] If the security assessment score is lower than the security threshold, the platform page is judged to be illegal.

[0023] Furthermore, the contextual behavior characteristics include at least environmental characteristics, user operation characteristics, timing characteristics and device characteristics.

[0024] Beneficial effect: The method of the present invention effectively combines the multimodal understanding characteristics of the AI scene recognition module, and can adapt to a variety of contextual behavior characteristics to calculate more accurate security assessment scores, making it more accurate in judging whether the platform-side page is legal.

[0025] Preferably, before step S10, the method of the present invention further comprises:

[0026] In response to the startup of the computer system, driving the virtual machine monitor to load the key interception module;

[0027] Use EPT (Extended Page Table) to divide sensitive memory areas and set the NX (non-executable) and UC (uncached) attributes of page table entries.

[0028] Furthermore, if a virtual machine is caught accessing a sensitive memory area, control of the virtual machine is transferred to the security monitoring module of the virtual machine monitor.

[0029] Beneficial effects: Compared with traditional plug-in and network-level password management, the present invention adopts security monitoring operations based on the system memory level, and uses single and asymmetric encryption at the memory level, which can effectively support password security management of multiple devices.

[0030] Preferably, before step S10, the method of the present invention further comprises:

[0031] Configure the virtual machine monitor to capture instructions for virtual machines accessing sensitive memory areas;

[0032] Among them, the instructions captured by the virtual machine to access sensitive memory areas at least include page table switching instructions, EPT invalidation instructions, field identification instructions, EPT violation instructions and CR register access instructions.

[0033] Preferably, after generating the short-term token and returning it to the platform front end, the method of the present invention further comprises:

[0034] Identify whether the short-term token has expired or has been used. If so, destroy the short-term token.

[0035] Beneficial Effects: The short-term tokens of this method automatically become invalid upon expiration. Furthermore, the corresponding token is immediately destroyed after each call is completed, ensuring a single-use token. Compared to traditional long-term credentials, short-term tokens reduce the attack surface while maintaining call efficiency, providing more secure call protection for sensitive operations.

[0036] In the second aspect, the present invention discloses a dynamic key protection system based on AI context perception, which includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the dynamic key protection method based on AI context perception recorded in the first aspect of the present invention is implemented.

[0037] The beneficial effects of the present invention are:

[0038] (1) Compared with the existing technology, the method of the present invention realizes credential encryption and abnormal operation interception, solving the problem of low security of existing password management technology.

[0039] (2) Compared with the existing technology, the method of the present invention mainly intercepts sensitive call behaviors in three directions: CPU execution of sensitive instructions, cryptographic library function call instructions, and abnormal context behavior tendencies. Among them, context behavior is identified in combination with the AI scene recognition module, realizing the effective combination of traditional technology and AI technology, which can more accurately intercept sensitive call behaviors and improve the security of password management technology.

[0040] (3) Compared with the existing technology and the traditional plug-in and network-level password management, the present invention adopts a security monitoring operation based on the system memory level, and uses a single and asymmetric encryption at the memory level, which can effectively support the password security management of multiple devices. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 Flowchart of the dynamic key protection method based on AI context perception in the first embodiment of the method of the present invention;

[0042] Figure 2 It is a schematic diagram of a dynamic key protection system based on AI context perception in the third embodiment of the method of the present invention. DETAILED DESCRIPTION

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work shall fall within the scope of protection of the present invention.

[0044] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] Example 1

[0046] like Figure 1 As shown, this embodiment discloses a dynamic key protection method based on AI context awareness, including:

[0047] S10: In response to the user inputting login credentials on the platform side, calling the preset AI scene recognition module to determine the legitimacy of the platform side page.

[0048] In this embodiment, the login credentials primarily refer to the user's account and password for logging into the platform. The platform can be any of WeChat, Taobao, JD.com, or an internal system management system. The AI scene recognition module utilizes a deep learning model, which can be based on a Transformer or lightweight CNN framework.

[0049] In other embodiments, the AI scene recognition module may also be replaced by a module that is a hybrid of a lightweight rule engine and AI.

[0050] S20: If the platform page is legal, generate a short-term token and return it to the platform front end.

[0051] In this embodiment, the short-term token refers to an identity authentication token.

[0052] In any of steps S10 to S20, the method of this embodiment performs the following steps in parallel:

[0053] Identify sensitive call behaviors and call the preset key interception module to intercept sensitive call behaviors.

[0054] Through the above technical solution, after the user enters their login credentials (account number and password) on the platform, if the platform page is legitimate, a short-term token is generated and returned to the platform front-end to complete the login. Compared to existing technologies, this short-term token is not a plaintext key / credential and cannot be directly obtained by the user. This short-term token effectively prevents plaintext leakage, debugging, and screenshot attacks. In addition, during the process of generating the short-term token, the method of this embodiment can also identify sensitive call behaviors and perform low-level interception on them.

[0055] Compared with the existing technology, the method of this embodiment realizes credential encryption and abnormal operation interception, solving the problem of low security of existing password management technology.

[0056] Furthermore, the steps of identifying sensitive call behaviors and calling a preset key interception module to intercept the sensitive call behaviors include:

[0057] Identify whether the CPU has a tendency to execute sensitive instructions, cryptographic library function call instructions, or abnormal context behavior.

[0058] If so, the preset key interception module is called to intercept the CPU from executing sensitive instructions, cryptography library function call instructions, or execution processes with abnormal context behavior tendencies.

[0059] Specifically, abnormal contextual behavior tendencies are identified using the AI scenario recognition module. The above-mentioned key interception module adopts a layered architecture design, which designs a three-level interception architecture to achieve full-link monitoring of sensitive operations, namely: Hardware layer interception: Capture sensitive instructions (such as memory operation instructions MOV[CR3], RAX) through CPU virtualization extensions (such as Intel VT-x's VMX operation) to trigger VMExit events.

[0060] System call layer interception: Replace the client's system call table at the hypervisor layer and redirect cryptographic-related APIs (such as CryptAcquireContext) to the security module.

[0061] Behavior analysis layer interception: The AI scenario recognition module analyzes the virtual machine process behavior sequence in real time and identifies abnormal key call patterns (such as batch export of keys during non-working hours).

[0062] For example, the sensitive discrimination operations based on the three-level interception architecture are shown in Table 1:

[0063] Table 1

[0064]

[0065] It should be noted that the above Table 1 is only an example description, and in specific application scenarios, the above exemplary features can be adaptively adjusted.

[0066] Through the above-mentioned technical solution and the architectural design of the key interception module, the method of this embodiment combines AI technology to achieve effective interception in three directions: CPU execution of sensitive instructions, cryptographic library function call instructions, and abnormal context behavior tendencies. It can more accurately intercept sensitive call behaviors and further improve the security of password management.

[0067] Preferably, in order to configure a good operating environment for the key interception module, before step S10, the method of this embodiment further includes:

[0068] S100: In response to startup of the computer system, driving a virtual machine monitor (Hypervisor) to load a key interception module.

[0069] In this embodiment, at the kernel driver or virtualization monitoring layer of the computer system, all password / key loading and calling operations are intercepted into a confidential area and protected by hardware-accelerated AES-GCM or TEE (such as Intel SGX) symmetric encryption.

[0070] S200: Use EPT to divide sensitive memory areas and set the NX and UC attributes of page table entries.

[0071] In addition to the configuration of steps S100-S200 above, the configuration process of the method of this embodiment also includes:

[0072] S300: Configuring a virtual machine monitor to capture instructions for a virtual machine to access a sensitive memory area.

[0073] Among them, the instructions captured by the virtual machine to access sensitive memory areas at least include page table switching instructions, EPT invalidation instructions, field identification instructions, EPT violation instructions and CR register access instructions.

[0074] Specifically, an exception interception flag is set in the VMCS (virtual machine control structure) to monitor privileged instructions such as MOV CR3 (page table switch) and INVEPT (EPT invalidation).

[0075] The instruction type is identified by the VMX_EXIT_REASON field, for example:

[0076] VM_EXIT_REASON_EPT_VIOLATION (EPT violation);

[0077] VM_EXIT_REASON_CR_ACCESS (CR register access).

[0078] Furthermore, if a virtual machine is caught accessing a sensitive memory area, control of the virtual machine is transferred to the security monitoring module of the virtual machine monitor.

[0079] Through the above design, the hypervisor ensures the security of each virtual machine through memory isolation. For example, Hyper-V manages partition physical memory through EPT, allowing non-privileged partitions to access only their own memory. Any out-of-bounds access triggers a VMExit, which the hypervisor can intercept and handle. Furthermore, the hypervisor can replace the client operating system's system call table or key APIs, switching execution to a secure path controlled by the hypervisor when cryptographic operations are called. To implement security monitoring within the hypervisor, memory pages requiring protection can be marked when the virtual machine is loaded, preventing unauthorized modification of the page table.

[0080] Therefore, the hypervisor implements deep security intervention in the virtualization layer, with the functions of transparently intercepting sensitive calls, isolating memory space, and encrypting data transmission, which can ensure that malicious kernels or virtual machines cannot steal keys after escaping.

[0081] In other embodiments, the above method can also be extended to container / Serverless environments to achieve cloud-native key protection.

[0082] Furthermore, sensitive data must always be encrypted during key transmission and storage. This method uses symmetric algorithms such as AES-GCM to encrypt passwords / keys at the hypervisor layer. The encryption key can be stored in a TEE (secure enclave) or secure hardware. When the host operating system or hypervisor accesses the virtual machine's memory, it can only see the encrypted data and cannot obtain the plaintext. The plaintext key only exists in the transient state of the CPU register. For example, it is loaded into the register when a critical instruction is called and immediately cleared or encrypted for storage after processing. Similar to AMD SEV-ES technology, the CPU register contents are encrypted when the virtual machine is suspended, further ensuring that the plaintext key will not be leaked in memory. Even if the hypervisor has permission, it can only access the encrypted ciphertext.

[0083] Compared to existing technologies, this invention utilizes security monitoring operations based on the system memory layer, employing single-shot, asymmetric encryption at the memory layer, effectively supporting password security management across multiple devices. This embodiment method is more sensitive to sensitive call behavior and offers enhanced security.

[0084] Furthermore, after generating a short-term token and returning it to the platform front end in step S20, the method of this embodiment further includes:

[0085] Identify whether the short-term token has expired or has been used. If so, destroy the short-term token.

[0086] Through the above technical solution, after verifying the legitimacy of the call, the method of this embodiment replaces the plaintext key with a short-term session token for subsequent access. The token has its own validity period (configurable, such as 60 seconds or 5 minutes), and automatically expires after expiration without additional operation. The validity period setting takes into account both security and practicality, significantly reducing the risk of long-term abuse of keys. The token automatically expires after expiration, and the corresponding token is immediately destroyed after each call is completed. The token is only used once. By monitoring the use of tokens in real time, the relevant tokens are immediately revoked and an alarm is issued for any abnormal behavior detected (such as token replay or cross-application use). Compared with traditional long-term credentials, short-term tokens reduce the attack surface while maintaining call efficiency, providing more secure call protection for sensitive operations.

[0087] For example, the logic for generating a short-term token in the method of this embodiment is as follows:

[0088] 1. Data binding: bind the token to the domain name of the current session (such as api.payment.com) and the operation type (decryption only);

[0089] 2. Signature mechanism: Use the hypervisor root key to perform HMAC-SHA256 signing on the token to generate an unalterable string;

[0090] Token=HMAC(SecretKey, UserID+Timestamp+Domain);

[0091] 3. Validity period control: Embed a timestamp in a short-term token (e.g., validity period of 60 seconds), and automatically trigger the revocation mechanism after expiration.

[0092] It is important to further explain that in this embodiment, short-lived tokens are generated with a key ID, a timestamp, and a random number, and are digitally signed (or HMACed) using a master key to ensure integrity. Each short-lived token is bound to a specific access context, including the domain name / application identifiers allowed access and operational permissions (such as read-only or decryption). The signing and binding policy ensure that the token can only be used within the specified environment and cannot be exported. This embodiment method verifies the legitimacy of the token within the hypervisor or TEE, thereby achieving refined access control.

[0093] Furthermore, to facilitate auditing and log analysis, all token acquisition and usage operations are recorded in the security log. Combined with the intelligent log analysis module, abnormal behaviors (such as unexpected process access or abnormal frequency) are detected in real time and alarms are triggered.

[0094] Based on the above technical description, the method of this embodiment has at least the following beneficial effects:

[0095] 1. Undetectable interception is achieved at the hypervisor layer. The key is stored in encrypted form throughout the process and is only briefly displayed in plain text in the CPU decryption register.

[0096] 2. When an attacker bypasses system call interception through ROP chaining, EPT violation detection at the hardware layer can intercept illegal memory access for a second time.

[0097] 3. Short-term tokens have an anti-replay mechanism. Each short-term token contains a random number Nonce and a session chain hash to prevent cross-session replay.

[0098] 4. Comprehensive encryption protection at the memory level is achieved, and plaintext credentials are only briefly decrypted in the CPU register, which can effectively prevent any form of memory leakage.

[0099] 5. A short-term token mechanism is used to replace static passwords / API keys. The short-term token has a short life cycle and its usage scope can be finely controlled.

[0100] 6. End-to-end auditing and protection are achieved, forming a closed loop from call interception, scenario identification to token management and log analysis, providing full-process security protection.

[0101] Example 2

[0102] Based on the first embodiment, this embodiment mainly optimizes AI scene recognition. This embodiment discloses a dynamic key protection method based on AI context perception, in which a preset AI scene recognition module is called to determine the legitimacy of the platform-side page, including:

[0103] S11: Obtain contextual behavior features.

[0104] The above-mentioned contextual behavior characteristics at least include environmental characteristics, user operation characteristics, time sequence characteristics and device characteristics.

[0105] More specifically, the following is an exemplary description of contextual behavior features:

[0106] Environmental characteristics: current window title (such as "Payment Platform-Login"), process path (C:\Program Files\Browser\chrome.exe).

[0107] Operational characteristics: API call sequence (CryptGenRandom→CryptProtectData).

[0108] Timing characteristics: operation frequency (e.g., 5 key requests per second triggering risk control).

[0109] Device characteristics: hardware fingerprint (TPM chip serial number), network environment (IP geographic location), facial features, or multi-factor authentication features.

[0110] Specifically, in the process of acquiring contextual behavioral features, machine learning is used to intelligently perceive the current calling environment. Input features include, but are not limited to, the active window name, process name and parent process information, the active window's UI element hierarchy (such as control type, input box identifier, button location, etc.), the current web page URL, and network access characteristics. Referring to context-aware authentication, the integration of environmental and operational context information can improve security and convenience. For example, when a user enters a password on a payment page, the browser window URL and UI structure provide a legitimacy prompt; however, when a non-browser or unknown window requests a key, it can be considered an abnormal context, raising vigilance.

[0111] S12: Encode and fuse the context behavior features to obtain a fusion result.

[0112] Specifically, to balance real-time performance and performance, the AI scene recognition module of the present invention uses a lightweight neural network model, such as a streamlined version of the Transformer or a small convolutional neural network (CNN). The AI scene recognition module first encodes discrete text features (such as window names, URLs, and process names) and fuses them with structured numerical features (such as network data statistics). Graph embedding methods or interface screenshots can be used as input for the UI hierarchy. Features are extracted through CNN. The Transformer architecture uses a self-attention mechanism to weightedly fuse each feature to generate a context-aware representation vector. For resource-constrained scenarios, technologies such as knowledge distillation and pruning can be introduced to further compress the size of the AI model so that it can run efficiently on edge devices or in secure containers.

[0113] S13: Input the fusion result into the pre-trained AI scene recognition module and output the safety assessment score.

[0114] S14: If the security assessment score is higher than the security threshold, the platform page is judged to be legal.

[0115] S15: If the security assessment score is lower than the security threshold, the platform page is judged to be illegal.

[0116] Through the above technical solution, the trained model is fed with the input fusion results and outputs a security assessment score to comprehensively judge the legitimacy and risk level of the current password call request. Different strategies are then triggered based on the prediction results. When the score is above the security threshold, a temporary access token is issued. When the score is below the security threshold, the request is rejected and an alarm is issued or multi-factor verification is triggered. Context-aware authentication can significantly improve system adaptability and reliability. In addition, the system relying on the method of this embodiment can continuously monitor user behavior during operation and fine-tune the model online or dynamically adjust the threshold based on newly collected samples, thereby further improving the detection accuracy.

[0117] Through real-time AI judgment and policy switching, this embodiment achieves intelligent access control. Compared to existing technologies, this embodiment eliminates the need for manual rule configuration and automatically identifies diverse usage scenarios through deep learning models, reducing misjudgments and operational costs.

[0118] Preferably, the method of this embodiment can also adapt to the AI model update mechanism and increase the amount of training according to new scenarios to improve recognition accuracy.

[0119] Example 3

[0120] like Figure 2As shown, this embodiment discloses a dynamic key protection system based on AI context perception, which includes a processor and a memory, wherein the memory stores computer program instructions. When the computer program instructions are executed by the processor, the dynamic key protection method based on AI context perception recorded in Example 1 or Example 2 is implemented.

[0121] While this specification has shown and described several embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Numerous modifications, variations, and alternatives will occur to those skilled in the art without departing from the concept and spirit of the present invention. It should be understood that in practicing the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A dynamic key protection method based on AI context awareness, characterized in that: include: S10: In response to the user entering login credentials on the platform, calling a preset AI scene recognition module to determine the legitimacy of the platform page; S20: If the platform page is legal, generate a short-term token and return it to the platform front end; In any one of steps S10 to S20, the method performs the following steps in parallel: Identify sensitive call behaviors and call a preset key interception module to intercept the sensitive call behaviors; Before step S10, the method further includes: In response to starting the computer system, driving the virtual machine monitor to load the key interception module; Use EPT to divide sensitive memory areas and set the NX and UC attributes of page table entries; The key interception module adopts a layered architecture design, which designs a three-level interception architecture to achieve full-link monitoring of sensitive operations, namely: Hardware-layer interception: Capture sensitive instructions through CPU virtualization extensions and trigger VMExit events; System call layer interception: Replace the client's system call table at the hypervisor layer and redirect cryptographic-related APIs to the security module; Behavior analysis layer interception: The AI scenario recognition module analyzes the virtual machine process behavior sequence in real time and identifies abnormal key call patterns.

2. The dynamic key protection method based on AI context awareness according to claim 1 is characterized in that: Identifying sensitive call behaviors and calling a preset key interception module to intercept the sensitive call behaviors includes: Identify whether the CPU has a tendency to execute sensitive instructions, cryptographic library function call instructions, or abnormal context behavior; If so, the preset key interception module is called to intercept the CPU from executing sensitive instructions, cryptography library function call instructions, or execution processes with abnormal context behavior tendencies.

3. The dynamic key protection method based on AI context awareness according to claim 2 is characterized in that: The abnormal tendency of contextual behavior is identified by the AI scene recognition module.

4. The dynamic key protection method based on AI context awareness according to claim 3 is characterized in that: Call the preset AI scene recognition module to determine the legitimacy of the platform page, including: Obtain contextual behavior features; Encoding and fusing the context behavior features to obtain a fusion result; Input the fusion result into a pre-trained AI scene recognition module and output a safety assessment score; If the security assessment score is higher than the security threshold, the platform page is judged to be legal; If the security assessment score is lower than the security threshold, the platform-side page is determined to be illegal.

5. The dynamic key protection method based on AI context awareness according to claim 4 is characterized in that: The contextual behavior characteristics include at least environmental characteristics, user operation characteristics, time sequence characteristics and device characteristics.

6. The dynamic key protection method based on AI context awareness according to claim 1 is characterized in that: If a virtual machine is captured accessing the sensitive memory area, the control right of the virtual machine is transferred to the security monitoring module of the virtual machine monitor.

7. The dynamic key protection method based on AI context awareness according to claim 1 is characterized in that: Before step S10, the method further includes: Configuring the virtual machine monitor to capture instructions for the virtual machine to access the sensitive memory area; The instructions for capturing the virtual machine to access the sensitive memory area include at least page table switching instructions, EPT invalidation instructions, field identification instructions, EPT violation instructions and CR register access instructions.

8. The dynamic key protection method based on AI context awareness according to claim 1 is characterized in that: After generating the short-term token and returning it to the platform front end, the method further includes: Identify whether the short-term token has expired or has been used, and if so, destroy the short-term token.

9. A dynamic key protection system based on AI context awareness, characterized in that: It includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the dynamic key protection method based on AI context perception according to any one of claims 1 to 8 is implemented.

Images