Privacy protection in large model access process

Abstract

A privacy protection method in a large model access process. A secure terminal is constructed by deploying a security aspect in a user terminal, and the security aspect can identify privacy data in an access request of the user terminal for accessing a target large model, and perform masking processing on the privacy data by means of aspect injection, so that the privacy data of a user is protected.

Description

Privacy protection during access to large models Technical Field

[0001] The present invention relates to the field of privacy protection technology, and in particular to a method and a user terminal for privacy protection during large model access. Background Art

[0002] Currently, when users interact with large models, the data they interact with may contain various sensitive personal or corporate information, such as PII and corporate keys. This can lead to the leakage of confidential data during interaction with large models. To address this issue, the industry urgently needs an effective user privacy protection solution. Summary of the Invention

[0003] One or more embodiments of this specification provide a privacy protection method and a user terminal during access to a large model, which can protect user privacy data during access to the large model by the user terminal.

[0004] In a first aspect, a privacy protection method during large model access is provided, comprising: deploying a security aspect on a user terminal; determining an access channel for the user terminal to access a target large model; in response to the security aspect intercepting an access request for accessing the target large model issued by the user terminal through the access channel, identifying whether sensitive data exists in the access request; if the sensitive data exists in the access request, desensitizing the sensitive data by the security aspect, and sending the desensitized data to the target large model.

[0005] As an optional implementation of the method described in the first aspect, the security aspect intercepts the access request for accessing the target large model issued by the user terminal through the access channel, specifically including: monitoring the access request issued by the user terminal through the access channel through the security aspect; if the access address in the access request is consistent with the access address of the target large model, intercepting the access request; if the access address in the access request is inconsistent with the access address of the target large model, releasing the access request.

[0006] As an optional implementation of the method described in the first aspect, identifying whether sensitive data exists in the access request may specifically include: identifying whether a preset type of sensitive data exists in the access request through regular matching.

[0007] As an optional implementation of the method described in the first aspect, the security aspect desensitizes the sensitive data, which may specifically include: the security aspect desensitizes the sensitive data according to the type of the sensitive data and a preset desensitization strategy for the type.

[0008] More specifically, the security aspect desensitizes the sensitive data, which may specifically include: the security aspect injects desensitizing code of the desensitizing policy into the access channel, and desensitizes the sensitive data by running the desensitizing code.

[0009] In a second aspect, a user terminal is provided, in which a security aspect is deployed. The security aspect is used to intercept the access request of the user terminal when the user terminal accesses a target large model through a preset access channel, and to identify whether there is sensitive data in the access request; if the sensitive data exists in the access request, the security aspect desensitizes the sensitive data and sends the desensitized data to the target large model.

[0010] As an optional implementation of the user terminal described in the second aspect, the security aspect is specifically used to monitor the access request issued by the user terminal through the access channel; if the access address in the access request is consistent with the access address of the target large model, the access request is intercepted; if the access address in the access request is inconsistent with the access address of the target large model, the access request is released.

[0011] As an optional implementation of the user terminal described in the second aspect, the security aspect is specifically used to identify whether there is sensitive data of a preset type in the access request through regular matching.

[0012] As an optional implementation of the user terminal described in the second aspect, the security aspect is specifically used to desensitize the sensitive data according to the type of the sensitive data and a preset desensitization policy for the type.

[0013] More specifically, in the user terminal, the security aspect desensitizes the sensitive data, specifically including: injecting desensitizing code of the desensitizing strategy into the access channel, and desensitizing the sensitive data by running the desensitizing code.

[0014] In a third aspect, a communication system is provided, which includes a large model server and the above-mentioned user terminal.

[0015] The beneficial effect of the privacy protection method during the large model access process described in one or more embodiments of this specification is that by deploying security aspects in the user terminal to build a security terminal, the security aspect can identify the privacy data in the access request of the user terminal to access the target large model, and desensitize the privacy data through aspect injection, thereby protecting the user's privacy data.

[0016] The user terminal and communication system described in the embodiments of this specification also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] FIG1 is a flow chart of a privacy protection method during access to a large model provided by one or more embodiments of this specification.

[0019] FIG2 is a schematic diagram of the structure of a communication system provided by one or more embodiments of this specification.

[0020] FIG3 is a schematic diagram of the structure of a user terminal provided in one or more embodiments of this specification.

[0021] FIG4 is a schematic structural diagram of an electronic device provided in one or more embodiments of this specification. DETAILED DESCRIPTION

[0022] Currently, large language models such as ChatGPT are rapidly developing, and a large number of users are using these cloud-based services for purposes such as code development and document writing. When using large cloud-based language models, users inevitably submit sensitive personal or corporate information, such as PII and corporate encryption keys. This behavior can lead to the leakage of confidential corporate or personal data, resulting in losses for the companies or individuals.

[0023] Typically, to prevent confidential data from being leaked, the industry builds an enterprise-level gateway or proxy server. This prevents users from directly accessing large model services within the enterprise and requires them to go through the gateway or proxy server. However, this method becomes unusable once the user leaves the coverage area of ​​the enterprise-level gateway or proxy server.

[0024] In view of this, one or more embodiments of this specification propose a privacy protection method and user terminal during large model access, which can effectively protect privacy data during the user terminal access to the large model without being restricted by the application site.

[0025] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments in this specification without creative work should fall within the scope of protection of this specification.

[0026] It should be noted that in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the method may include more or fewer steps than those described in this specification. In addition, a single step described in this specification may be broken down into multiple steps for description in other embodiments, and multiple steps described in this specification may be combined into a single step for description in other embodiments.

[0027] Those skilled in the art will appreciate that the terms used in the embodiments of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The singular forms "a," "an," "the," and "the" used in the embodiments of the present invention and the appended claims are intended to include the plural forms, unless the context clearly indicates otherwise.

[0028] In some embodiments, a privacy protection method during large model access (hereinafter referred to as the privacy protection method) is provided. Referring to FIG. 1 , the method includes steps S100 - S106 .

[0029] S100: Deploy security aspects on user terminals.

[0030] The aforementioned security aspects are a security system concept. Its core concept is to decouple security management from business logic by embedding security control points (pointcuts) at every level and aspect of business logic. These aspects are independent of the main business logic and do not affect the implementation of primary functions. This allows for the addition of additional security checks and protection mechanisms without changing the original business logic, improving the overall security of the software system. This security aspect system uses AOP (Aspect-Oriented Programming) to insert security aspects between atomic functions and the business logic layer. This enables behavioral monitoring and control of functions within the application at runtime, providing risk discovery and governance capabilities for privacy-preserving businesses.

[0031] The above-mentioned user terminal is an intelligent terminal that can access the target large model and deploy the above-mentioned security aspects, such as a desktop computer, notebook, Pad, smart phone, etc.

[0032] S102: Determine an access channel for the user terminal to access the target large model.

[0033] The above access channels refer to common ways for users to access big model services, such as big model client apps, Python scripts, browsers, and big model plug-ins for various development tools.

[0034] S104: In response to the security aspect intercepting an access request for accessing the target large model sent by the user terminal through the access channel, identifying whether sensitive data is included in the access request.

[0035] The security aspect described above monitors all access channels identified in step S102. That is, whenever a user terminal sends an access request through one of these access channels, the security aspect intercepts the request and determines whether it is intended to access the target large model. If so, the security aspect then identifies whether the request contains sensitive data.

[0036] In some possible implementations, the security aspect can identify whether the access request is for accessing the target large model by identifying the access address in the access request. For example, for a certain cloud-based large model service, the access request can be identified as using the http / https protocol. If the identified access domain name is the domain name or URL of the cloud service for the large model, the access request is determined to be for accessing the large model, and the security aspect intercepts the access request. If the identified access domain name is not the domain name or URL of the cloud service for the large model, the access request is determined not to be for accessing the large model, and the security aspect allows the access request.

[0037] For access requests to the target large model issued by user terminals through access channels intercepted by the security aspect, it is also necessary to identify whether sensitive data exists in the access request. In some feasible implementations, regular matching can be used to identify whether preset types of sensitive data exist in the access request. For example, the types of sensitive data can be personal privacy information and enterprise sensitive data. Among them, personal privacy can be personal identification information PII, which refers to a direct identifier that can uniquely identify an individual when used alone or with other relevant data (such as passport information), or a quasi-identifier (such as ethnic group) that can be combined with other quasi-identifiers (such as date of birth) to successfully identify an individual. Common PII include: mobile phone number, ID number, driver's license, mailing address, etc. Enterprise sensitive data can be keys, authentication credentials, etc. For personal privacy information and enterprise sensitive data, regular expressions can be set separately to describe specific sensitive data. Therefore, when identifying whether sensitive data exists in the access request, matching can be performed through pre-set regular expressions.

[0038] S106: If there is sensitive data in the access request, the security aspect desensitizes the sensitive data and sends the desensitized data to the target large model.

[0039] In some feasible implementations, sensitive data can be desensitized according to a pre-set desensitization strategy based on the type of sensitive data. For example, desensitization code for the desensitization strategy can be injected into the access channel, and the corresponding sensitive data can be desensitized by running the desensitization code. The above-mentioned desensitization strategy may include deleting content, masking certain content, etc.

[0040] As can be seen from the privacy protection method provided above, this method uses aspect technology to inject desensitizing code snippets into the common methods used by user terminals to access the target large model service. This desensitizes sensitive data, such as personal privacy information (such as PII) and sensitive enterprise data (such as keys and authentication credentials), within the content of requests sent to the target large model. This privacy protection method does not require the establishment of any additional gateways or proxy devices in the application environment, does not require changes to the way users access the large model service, and its use is not restricted by any location.

[0041] Please refer to Figure 2, which shows a communication system that applies the above-mentioned privacy protection method. It should be noted that the privacy protection method during the large model access process described in one or more embodiments of this application can be applied to the communication system shown in Figure 1, but is not limited to this communication system.

[0042] As shown in FIG2 , the communication system includes a user terminal 20 and a large model server 21. The user terminal 20 establishes a communication connection with the large model server 21 via a communication link, and the communication link can be a wired network or a wireless network. For example, the user terminal 20 can establish a communication connection with the large model server 21 using a communication method such as WIFI, Bluetooth, infrared, etc. Alternatively, the user terminal 20 can also establish a communication connection with the large model server 21 via a mobile network, wherein the network standard of the mobile network can be any one of 2G (GSM), 2.5G (GPRS), 3G (WCDMA, TD-SCDMA, CDMA2000, UTMS), 4G (LTE), 4G+ (LTE+), WiMax, etc.

[0043] The user terminal 20 can be a mobile terminal, such as a mobile phone, notebook, iPad, etc. A security aspect is deployed in the user terminal 20. The security aspect is used to monitor the access requests sent by the user terminal 20 in the preset access channel and identify whether these access requests are for accessing the target large model. When an access request is identified as accessing the target large model, the security aspect intercepts the access request and identifies whether there is sensitive data in the access request. If there is sensitive data in the access request, the security aspect desensitizes the identified sensitive data and sends the desensitized data to the target large model. If there is no sensitive data in the access request, the security aspect sends the access request directly to the target large model.

[0044] The large model server 21 is a server of a large model service provider, providing large model services to users. The large model server 21 can be any device, equipment, platform, or device cluster with computing and processing capabilities. In this embodiment, the implementation of the large model server 21 is not limited. For example, the large model server 21 can be a single server or a server cluster composed of multiple servers. The large model server 21 can also be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system.

[0045] When the above communication system applies the privacy protection method during the large model access process described in this embodiment, the user terminal 20 executes the following process.

[0046] S200: Determine the access channel for the user terminal 20 to access the large model server 21.

[0047] S202: Monitor the access channel mentioned above, intercept the access request to the large model server 21 sent by the user terminal 20 through the access channel through the security aspect, and identify whether there is sensitive data in the access request.

[0048] S204: If there is sensitive data in the access request, the security aspect desensitizes the sensitive data and sends the desensitized data to the large model server 21.

[0049] S206: If the access request does not contain sensitive data, the security aspect releases the access request.

[0050] Corresponding to the above-mentioned communication system and privacy protection method, this embodiment also provides a user terminal 20, which is deployed with a security aspect. The security aspect is used to intercept the access request of the user terminal 20 when the user terminal 20 accesses the target large model through a preset access channel, and identify whether there is sensitive data in the access request; if there is sensitive data in the access request, the security aspect desensitizes the sensitive data and sends the desensitized data to the target large model.

[0051] In some feasible implementations, the security aspect deployed in the above-mentioned user terminal 20 is specifically used to monitor access requests issued by the user terminal through access channels. Since the access addresses of these third-party large models are usually public, it is possible to identify whether the intercepted request is for accessing the target large model by matching the access address. If the access address in the access request is consistent with the access address of the target large model, the access request is intercepted; if the access address in the access request is inconsistent with the access address of the target large model, the access request is released. For example, for a certain cloud-based large model service, access using the http / https protocol can be identified from the access request. If the identified access domain name is the domain name or url of the cloud service of the large model, it is determined that the access request is for accessing the large model. At this time, the security aspect intercepts the access request. If the identified access domain name is not the domain name or url of the cloud service of the large model, it is determined that the access request is not for accessing the large model. At this time, the security aspect releases the access request.

[0052] In some feasible implementations, the security aspect deployed in the above-mentioned user terminal 20 can use a regular matching method to identify whether there is sensitive data in the access request. For example, the types of sensitive data can be personal privacy information and enterprise sensitive data. Among them, personal privacy can be personal identification information PII, which refers to a direct identifier (such as passport information) that can uniquely identify an individual when used alone or with other relevant data, or a quasi-identifier (such as ethnic group) that can be combined with other quasi-identifiers (such as date of birth) to successfully identify an individual. Common PII include: mobile phone number, ID number, driver's license, mailing address, etc. Enterprise sensitive data can be keys, authentication credentials, etc. For personal privacy information and enterprise sensitive data, regular expressions can be set separately to describe specific sensitive data. Therefore, when identifying whether there is sensitive data in the access request, matching can be performed through pre-set regular expressions.

[0053] In some feasible implementations, the security aspect deployed in the user terminal 20 can desensitize sensitive data according to the type of sensitive data and a preset desensitization strategy for that type. For example, desensitization code of the desensitization strategy can be injected into the access channel, and the corresponding sensitive data can be desensitized by running the desensitization code. The above-mentioned desensitization strategy can include deleting content, masking certain content, etc.

[0054] Please refer to Figure 3, which shows a schematic structural diagram of the above-mentioned user terminal 20. As shown in Figure 3, the user terminal includes: a security aspect base 31 and a desensitization module 32 deployed in the security aspect base.

[0055] The security aspect base 31 is used to monitor the access channels commonly used by the user terminal 20. When it monitors the access request of the user terminal 20 to access the target large model, it identifies whether there is sensitive data in the access request. If there is sensitive data in the access request, the desensitization strategy is determined according to the type of sensitive data, and the desensitization strategy information is sent to the desensitization module 32.

[0056] The desensitizing module 32 can pre-store code snippets for different types of desensitizing strategies. After receiving the desensitizing strategy information issued by the security aspect base 31, it can select the corresponding desensitizing strategy code snippet according to the desensitizing strategy information, and then inject these desensitizing strategy code snippets into the corresponding tangent point according to the tangent point information in the desensitizing strategy information to realize desensitizing processing of specific data.

[0057] For the above-mentioned user terminal 20, taking a module as an example of a software functional unit, the security aspect base 31 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, and a container. Furthermore, the above-mentioned computing instance may be one or more. For example, the security aspect base 31 may include code running on multiple hosts / virtual machines / containers. The multiple hosts / virtual machines / containers used to run the code may be distributed in the same region or in different regions. Furthermore, the multiple hosts / virtual machines / containers used to run the code may be distributed in the same availability zone (AZ) or in different AZs, each AZ including one data center or multiple geographically close data centers. Typically, a region may include multiple AZs.

[0058] Similarly, multiple hosts / virtual machines / containers running the code can be distributed within the same virtual private cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Cross-region communication between two VPCs within the same region, or between VPCs in different regions, requires a communication gateway within each VPC to interconnect the VPCs.

[0059] As an example of a hardware functional unit, the security aspect base 31 may include at least one computing device, such as a server. Alternatively, the security aspect base 31 may be implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL), or any combination thereof.

[0060] The multiple computing devices included in the security aspect base 31 can be distributed in the same region or in different regions. The multiple computing devices included in the security aspect base 31 can be distributed in the same AZ or in different AZs. Similarly, the multiple computing devices included in the security aspect base 31 can be distributed in the same VPC or in multiple VPCs. The multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0061] In this implementation, the security aspect base 31 can also be applied to computing devices such as computers and servers, or to a computing device cluster including at least one computing device, to implement privacy protection during large model access.

[0062] In some embodiments, an electronic device is also provided for implementing the user terminal 20 described above. Referring to Figure 4 , the electronic device includes a bus 401, a processor 402, a memory 403, and a communication interface 404. The processor 402, the memory 403, and the communication interface 404 communicate with each other via the bus 401. The electronic device may be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in the electronic device.

[0063] Bus 401 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, control buses, and the like. For ease of illustration, FIG4 illustrates a single bus line, but this does not imply a single bus or type of bus. Bus 401 may include a path for transmitting information between various components of an electronic device (e.g., processor 402, memory 403, and communication interface 404).

[0064] The processor 402 may include any one or more processors such as a CPU, a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0065] The memory 403 may include a volatile memory, such as a random access memory (RAM). The memory 403 may also include a non-volatile memory, such as a read-only memory (ROM), a flash memory, a hard disk drive (HDD), or a solid state drive (SSD).

[0066] The memory 403 stores executable program codes, and the processor 402 executes the executable program codes to implement the aforementioned functions of the security terminal 20 , that is, to implement the aforementioned privacy protection method during the large model access process.

[0067] The communication interface 404 uses a transceiver module such as, but not limited to, a network interface card or a transceiver to implement communication between the electronic device and other devices or a communication network.

[0068] In some embodiments, a computer-readable storage medium is further provided, on which a computer program is stored. When the computer program is executed by a processor, the functions of the above-mentioned security aspect are implemented.

[0069] The computer-readable storage medium can be any available medium that can be stored by the electronic device or a data storage device such as a data center that contains one or more available media. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive). The computer-readable storage medium includes instructions that instruct the electronic device to execute the model training method.

[0070] It should be understood that the structures illustrated in the embodiments of this specification do not constitute specific limitations on the systems of the embodiments of this specification. In other embodiments of the specification, the above-mentioned system may include more or fewer components than shown in the figure, or some components may be combined, some components may be separated, or the components may be arranged differently. The components shown in the figure may be implemented in hardware, software, or a combination of software and hardware.

[0071] The various embodiments in this specification are described in a progressive manner. Similar parts between the various embodiments can be referred to in conjunction with each other. Each embodiment focuses on the differences from other embodiments. In particular, the device embodiments are generally similar to the method embodiments, so the description is relatively simple. For relevant parts, refer to the description of the method embodiments.

[0072] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0073] It should be noted that the above examples are merely specific embodiments of the present invention. Obviously, the present invention is not limited to the above examples, and many similar variations are possible. All variations directly derived from or associating with the present invention by those skilled in the art are intended to fall within the scope of protection of the present invention.

Claims

1. A privacy protection method during access to a large model, comprising: Deploy security aspects on user terminals; Determining an access channel for the user terminal to access the target large model; In response to the security aspect intercepting an access request for accessing the target large model sent by the user terminal through the access channel, identifying whether sensitive data exists in the access request; If the sensitive data exists in the access request, the security aspect desensitizes the sensitive data and sends the desensitized data to the target large model.

2. The method according to claim 1, wherein the security aspect intercepts the access request for accessing the target large model issued by the user terminal through the access channel, specifically comprising: monitoring, through the security aspect, an access request sent by the user terminal through the access channel; If the access address in the access request is consistent with the access address of the target large model, intercepting the access request; If the access address in the access request is inconsistent with the access address of the target large model, the access request is released.

3. The method of claim 1, wherein identifying whether sensitive data is present in the access request comprises: Identify whether there is sensitive data of a preset type in the access request by regular matching.

4. The method according to claim 1, wherein the security aspect performs desensitization processing on the sensitive data, specifically comprising: The security aspect desensitizes the sensitive data according to the type of the sensitive data and a preset desensitization strategy for the type.

5. The method according to claim 4, wherein the security aspect desensitizes the sensitive data, specifically comprising: The security aspect injects the desensitizing code of the desensitizing strategy into the access channel, and desensitizes the sensitive data by running the desensitizing code.

6. A user terminal is deployed with a security aspect, which is used to intercept the access request of the user terminal when the user terminal accesses a target large model through a preset access channel, and identify whether there is sensitive data in the access request; if the access request contains the sensitive data, the security aspect desensitizes the sensitive data and sends the desensitized data to the target large model.

7. The user terminal according to claim 6, wherein: The security aspect is specifically used to monitor the access request sent by the user terminal through the access channel; if the access address in the access request is consistent with the access address of the target large model, the access request is intercepted; If the access address in the access request is inconsistent with the access address of the target large model, the access request is released.

8. The user terminal according to claim 6, wherein: The security aspect is specifically used to identify whether there is sensitive data of a preset type in the access request through regular matching.

9. The user terminal according to claim 6, wherein: The security aspect is specifically used to desensitize the sensitive data according to the type of the sensitive data and a preset desensitization strategy for the type.

10. The user terminal according to claim 9, wherein: The security aspect desensitizes the sensitive data, specifically including: The desensitizing code of the desensitizing strategy is injected into the access channel, and the sensitive data is desensitized by running the desensitizing code.

11. A communication system comprising: A large model server, and a user terminal as claimed in any one of claims 6 to 10.

Images