Privacy protection AI decision system and method based on digital signature

Abstract

The invention discloses a privacy protection AI decision making system based on digital signature, comprising a security chip which is internally provided with a cryptographic hash function H and is used for generating an irreversible large Token digital asset package through operation Token = H (ID + Datahash + TrustLevel) according to a user unique identity ID, an original data hash value Datahash and a dynamic trust level TrustLevel; the digital signature management module is used for generating a multi-level digital signature system, binding a digital signature with the multi-dimensional organization information and carrying out digital signature on the large Token digital asset package; the privacy protection module is used for calling the digital signature management module to carry out signature verification and triggering a desensitization program; and the AI large model interaction module is used for receiving the signed and desensitized large Token digital asset package and generating a decision result by utilizing AI analysis capability.

Description

Technical Field

[0001] This invention belongs to the field of network security technology, specifically relating to digital identity authentication systems and methods. Background Technology

[0002] With the deep integration of Artificial Intelligence (AI) and big data technologies, various stakeholders, including governments, communities, businesses, families, and individuals, face increasingly severe structural and technological challenges when using private data for AI-driven decision-making. Existing technologies typically require uploading raw, sensitive data—Personally Identifiable Information (PII)—to large cloud-based models for inference.

[0003] In patent application CN 202511774541.4, Hu Jinqian disclosed a comprehensive integrated large-token data management system and method, proposing to aggregate and encapsulate digital asset packages into a single verifiable large-token data structure, where the token primary key, once generated, cannot be reversed to deduce the original identity. However, even with encrypted transmission, there is still a risk that the original sensitive information may be leaked during model interaction through reverse reasoning or training, making it impossible to truly achieve "data usable but invisible" and physical isolation of the original data from leaving the domain. Existing technologies for processing private data face the following pressing technical challenges: traditional data management systems lack authoritative identity verification mechanisms; during data transfer, AI receivers cannot accurately identify data ownership (such as government seals, corporate seals, personal seals, etc.), resulting in insufficient legal validity and traceability, making it difficult to form legally supported decision-making evidence; large AI models require significant computing power for preprocessing, cleaning, and semantic parsing when receiving unstructured raw data; and a large amount of redundant data details (such as thousands of financial transaction records or physiological indicators) not only increases the burden on network transmission but also significantly increases the inference latency and energy consumption of the receiver model.

[0004] Furthermore, in the context of global data flow, different jurisdictions have varying standards for digital signatures (including electronic seals and / or electronic signatures) and privacy protection. Foreign countries often employ signature encryption systems, which are difficult to integrate with China's domestic electronic seal system, leading to significant trust gaps and compliance costs when data is exchanged across borders and domains. The value of data assets changes in real time with the behavior of the subject (such as credit changes), and existing static data packages cannot reflect the dynamic trust level of the subject in real time. Moreover, the lack of an immutable evidence storage mechanism based on blockchain means that AI decisions may be based on outdated or tampered credit information.

[0005] Therefore, there is an urgent need for an efficient solution that can overcome the above-mentioned shortcomings.

[0006] It should be noted that the above description of the technical background is only for the purpose of providing a clear and complete explanation of the technical solutions of the present invention and facilitating understanding by those skilled in the art. It should not be assumed that the above technical solutions are known to those skilled in the art simply because they have been described in the background section of this invention. Summary of the Invention

[0007] The purpose of this invention is to overcome the shortcomings of the prior art and provide a privacy-protecting data ownership confirmation and AI decision-making system based on digital signatures. By encapsulating data into large tokens and binding them with digital signatures, the system achieves data usability without visibility and energy conservation and emission reduction.

[0008] This application discloses a privacy-preserving AI decision-making system based on digital signatures, including:

[0009] A security chip with a built-in cryptographic hash function H is used to calculate based on the user's unique identifier ID, the original data hash value Data_hash, and the dynamic trust level TrustLevel.

[0010] Token=H(ID+Data_hash+TrustLevel) (Formula 1)

[0011] Generate an irreversible large token digital asset package;

[0012] A digital signature management module is used to generate a multi-level digital signature system, bind digital signatures with multi-dimensional organizational information, and digitally sign large token digital asset packages.

[0013] A privacy protection module is used to call the aforementioned digital signature management module to verify digital signatures and trigger an anonymization process to ensure that the original information does not leave the domain; and

[0014] An AI large-scale model interaction module is used to receive signed and anonymized large-scale token digital asset packages, and use AI analysis capabilities to generate decision results. During the processing, no original sensitive privacy data is touched.

[0015] Furthermore, the aforementioned AI decision-making system for privacy protection based on digital signatures also includes a multilingual signature cross-domain adaptation module, used to achieve equivalent conversion and cross-border authentication between cross-domain electronic seals and digital signature standards. This module parses the multi-element authentication identifiers of the source digital signature, extracting metadata such as identity credential hash values, timestamps, and signature hierarchy relationships. Based on the target country's pre-defined compliance policy library, the module repackages this metadata into an X.509 digital certificate format conforming to the Public Key Infrastructure (PKI) standard, where the signature hierarchy relationship is mapped to custom fields in the certificate's extended domain. Finally, the converted digital signature is re-signed using an encryption algorithm recognized by the target country, generating an encrypted proof document with legal validity in the target jurisdiction.

[0016] Furthermore, the aforementioned large token digital asset package has a dynamic trust level, which is updated according to the user's credit behavior. The updated trust level is digitally signed and then packaged and uploaded to the blockchain network for notarization. The notarized content includes at least the user identifier hash value, the new trust level, the timestamp, and the signature information.

[0017] Furthermore, the aforementioned privacy protection module parses the unstructured raw files of multi-dimensional data into key attribute fields and only transmits the encapsulated structured digital asset package; it aggregates multiple raw detailed data into values ​​or labels that reflect statistical characteristics; and it injects default business semantic labels into the digital asset package to reduce the parsing burden on the receiving end's computing model.

[0018] Furthermore, the semantic tag injection program executed by the aforementioned security chip further includes: precisely aligning the de-identified structured data with the default business knowledge graph; wherein, the medical field alignment maps the original medical data to the medical knowledge graph, converting specific physiological values ​​into standardized clinical status tags; the financial field alignment maps the original bank statement data to the financial account code system, converting it into corresponding accounting account tags and statistical values; and the administrative field alignment maps the original data attributes to the administrative affiliation attribute graph, converting it into structured geographical location or administrative division tags.

[0019] Furthermore, the aforementioned digital signature management module performs signature verification. When a user triggers a certain function, the aforementioned privacy protection module will send a call request to the aforementioned security chip, requesting that the current operation instruction be digitally signed. The aforementioned digital signature management module initiates multi-factor authentication, and the aforementioned security chip uses its private key to encrypt the call request instruction, generating a temporary digital signature.

[0020] Furthermore, the aforementioned privacy protection module verifies the legitimacy of the temporary digital signature using a public key. After successful signature verification, the privacy protection module sends a de-identification instruction to the aforementioned security chip. The aforementioned security chip completes the structured de-identification operation within the chip and transmits the de-identified large token to the Trusted Execution Environment (TEE) via a secure channel for temporary storage and subsequent processing.

[0021] Furthermore, after successful signature verification, the aforementioned privacy protection module performs structured desensitization to transform the data, and directly removes personal identification information such as names and ID card numbers from the original data to remove identifiers. In addition, the aforementioned privacy protection module aligns the specific values ​​to the default knowledge graph to inject semantic tags.

[0022] Furthermore, in a family setting, the aforementioned privacy protection module will perform structured anonymization on the original family data, removing all members' names, ID numbers, and other personal identification information, and retaining only the relationships between members as metadata injected into the large token.

[0023] Furthermore, the aforementioned privacy protection module is also used to generate a zero-knowledge proof based on the original data for requests from the AI ​​big model that only require verification that the data meets specific conditions (such as credit score > threshold, age > threshold), and send the proof along with the signed and desensitized big token to the AI ​​big model interaction module, so as to prove to the AI ​​big model that the above conditions are met without disclosing the original values.

[0024] This application also discloses a privacy-preserving AI decision-making method based on digital signatures, comprising the following steps: S1: Receive user operation instructions and verify their identity and permissions through digital signatures; S2: Obtain the original data of the corresponding digital asset package from the security chip; S3: Confirm ownership of the original data and encapsulate it into a structured large token that hides specific organizational information; S4: De-identify the large token according to decision requirements and sign it using a digital signature to ensure the source is trustworthy; S5: Send the signed and de-identified large token to the AI ​​large model for multimodal collaborative analysis; S6: Receive and return the decision results generated by the AI ​​large model.

[0025] The present invention also discloses a computer-readable storage medium, characterized in that: the computer-readable storage medium includes a stored computer program, wherein, when the computer program is run by a processor, it controls the device where the storage medium is located to execute the above-mentioned privacy-preserving AI decision-making method based on digital signature.

[0026] The beneficial effects of this invention are as follows: Original data is strictly locked within a secure chip at the edge. Through structured desensitization technology, only hash-encrypted and semantically tagged large tokens are uploaded to the AI ​​large model. The processing flow complies with the anonymization requirements of relevant laws, completely eliminating the risk of sensitive information leakage in the cloud. This invention reduces the amount of data transmitted by 80% to 90% compared to the original files by performing data aggregation and structured transformation at the edge. In this invention, the AI ​​large model only needs to process structured data aligned with tags, eliminating complex data cleaning and preprocessing calculations, significantly reducing computing power consumption and carbon emissions. This invention binds multi-dimensional organizational information to data through digital signatures, transforming dispersed digital information into legally valid digital assets. The large token of this invention embeds a dynamic trust level that updates in real time with behavior, providing highly credible data weighting for AI decision-making.

[0027] This invention is not a simple aggregation of single technologies, but rather an organic integration of core technologies such as "large token encapsulation," "multi-dimensional digital signature binding," "structured anonymization," and "zero-knowledge proof," giving rise to a series of unprecedented innovative technological achievements that fundamentally reshape the circulation and application model of data elements. Traditionally, data value is significantly diminished once it is anonymized. This invention successfully breaks this paradox through structured anonymization technology. It transforms raw data into a new type of data asset—large token—that is irreversibly reproducible (ensuring privacy) yet rich in semantics and logic (preserving value). This new form allows data to be efficiently utilized by AI models while strictly protecting the privacy of organizations / individuals, releasing enormous economic and social value.

[0028] Existing electronic signatures only sign the entire document and cannot express the complex organizational relationships behind it (such as multiple legal entities managing a company, or authorization from family members). The multi-dimensional digital signature system of this invention can deeply bind multi-dimensional organizational information such as "who you are," "who you represent," and "what permissions you have" to the data content itself. This allows AI models to not only see the content when receiving data but also understand the legitimacy and authority of its source, providing richer context for intelligent decision-making.

[0029] This invention integrates dynamic trust level as a core parameter into the large token generation process, creating a positive incentive mechanism where "better credit leads to better service." This not only improves the efficiency of high-credit entities, but more importantly, it transforms the abstract concept of "social credit" into quantifiable, verifiable, and executable technical parameters, driving the entire digital society towards a more trustworthy and efficient governance model.

[0030] This invention shifts the focus of data processing from the cloud back to the user terminal. By completing data parsing, aggregation, and anonymization within a secure chip, it achieves a new paradigm of "data remains still, model moves lightly." This not only brings energy savings of 80%–90% but also reduces the cross-domain transmission of massive amounts of raw data at the source, greatly reducing the risk of network congestion and data leakage, and opening a green channel for the sustainable development of AI.

[0031] This invention, for the first time, incorporates five basic social units—government, community, enterprise, family, and individual—into a unified framework. Through a model of "government data packages, enterprise data packages, family data packages, individual data packages…", it constructs a truly real-world-centric digital identity and asset management system. This provides a robust, reliable, and scalable identity infrastructure for future smart cities, digital governments, and metaverse scenarios.

[0032] In summary, the innovative technical effect of this invention lies in the fact that it not only resolves the current contradiction between data security and AI applications, but also proactively defines a new standard for the safe, efficient, reliable, green, and globalized circulation of data elements. Its influence will transcend a single technical field and actively promote the transformation of the digital economy and social governance. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of a privacy-preserving AI decision-making system based on digital signature, according to an embodiment of the present invention.

[0034] Figure 2 This is a flowchart of a privacy-preserving AI decision-making method based on digital signature, according to an embodiment of the present invention.

[0035] The reference numerals in the above figures are as follows:

[0036] The privacy protection AI decision-making system based on digital signature 100, security chip 10, digital signature management module 20, privacy protection module 30, AI large model interaction module 40, multilingual signature cross-domain adaptation module 50, and steps S1 to S6 are as follows. Detailed Implementation

[0037] To better understand this invention, the following embodiments are provided in conjunction with the accompanying drawings. It should be understood that the embodiments of this invention are for illustrative purposes only and not for limiting the invention; the scope of protection of this invention is defined solely by the claims. The embodiments provided are merely preferred embodiments and are not intended to limit the invention in any way. Those skilled in the art can make changes, equivalent substitutions, or modifications based on the content of this invention to form different implementations. However, any changes and modifications, and any equivalent substitutions made to the method of this invention without departing from the inventive concept are within the scope of protection of this invention.

[0038] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0039] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms “comprising” and / or “including” are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0040] First, please refer to Figure 1 , Figure 1 This is a schematic diagram of a privacy-preserving AI decision-making system 100 based on digital signature, according to an embodiment of the present invention. Figure 1 As shown, the present invention discloses a privacy-preserving AI decision-making system 100 based on digital signatures, comprising a security chip 10, a digital signature management module 20, a privacy protection module 30, and an AI large-scale model interaction module 40. The security chip 10 is used to encapsulate multi-dimensional data from government, communities, enterprises, families, and individuals into an irreversible large-token digital asset package. The digital signature management module 20 is used to generate a multi-level digital signature system, bind digital signatures to multi-dimensional organizational information, and digitally sign the large-token.

[0041] It is worth noting that the aforementioned digital signature system includes one or more of the following: government special seal, enterprise multi-person collaborative seal, family joint seal, spousal joint management seal, and personal name seal. It is used to bind the identity, authority, and qualifications of the subject with the digital signature through multi-certificate, multi-encryption, multi-element, and multi-platform co-signing and mutual recognition technology, generate a unique authentication identifier, and realize the unified management of all digital asset packages by one seal.

[0042] In addition, the aforementioned privacy protection module 30 is used to call the digital signature for signature verification and trigger the desensitization procedure before submitting the large token to the AI ​​large model, ensuring that the original information does not leave the domain. The aforementioned AI large model interaction module 40 is used to receive the signed and desensitized large token, use AI analysis capabilities to generate decision results, and does not touch the original sensitive privacy data during the processing.

[0043] It is worth noting that the aforementioned large token of this invention has a dynamic trust level, which is updated according to the user's credit behavior. The updated trust level and signature will be uploaded to the blockchain network for evidence storage, ensuring the immutability and traceability of the credit change across the entire network.

[0044] In one embodiment of the present invention, the aforementioned AI decision-making system 100 based on digital signatures further includes a credit assessment module. The dynamic trust level is periodically calculated and generated by the credit assessment module based on the user's historical behavioral data (including but not limited to data usage compliance, identity authentication frequency, business performance records, etc.), with an update cycle of T (T≥1 day). The updated trust level, after being digitally signed, is packaged and uploaded to a blockchain network for evidence storage. The evidence storage content includes at least the user identifier hash value, the new trust level, a timestamp, and signature information.

[0045] The security chip of this invention recalculates the Token parameters based on a unique identifier, the original data hash value, and the dynamic trust level using a cryptographic hash function H (e.g., according to the People's Republic of China National Standard GB / T 32905-2016 Information Security Technology SM3 Cryptographic Hash Algorithm or the US National Institute of Standards and Technology's cryptographic hash function SHA-3 (Secure Hash Algorithm 3)).

[0046] Token=H(ID+Data_hash+TrustLevel) (Formula 1)

[0047] In the formula, ID is a unique identifier, Data_hash is the original data hash value, and TrustLevel is the dynamic trust level. The ID, the unique identifier, is a globally unique ID within the system, issued by an authoritative institution and securely stored in the aforementioned security chip 10. The original data hash value (Data_hash) can be, for example, the security chip 10 reading the scanned PDF inspection report and performing a hash operation (e.g., using the SHA-256 algorithm) on its entire binary content to obtain a fixed-length digest. This digest represents the report's digital fingerprint; any minor modification will cause a drastic change in the hash value. The dynamic trust level, TrustLevel, is a score that the system evaluates in real time based on past credit behavior (such as timely tax payments, no illegal records, and compliant data usage). This score is periodically updated and securely stored within the aforementioned security chip 10.

[0048] For example, the aforementioned security chip 10 concatenates the three parameters into a single input string using a specific delimiter. Then, the security chip 10 uses a cryptographic hash function H to operate on the input string to obtain another unique hash value. Due to the avalanche effect of hash function H, even if the TrustLevel changes from "AA" to "AAA", the final generated token will be completely different. Furthermore, no information regarding the user's identity, data content, or credit status can be deduced from the token, thus protecting user privacy. This invention successfully binds user identity, data content, and credit status tightly together; tampering with any one of these elements will cause token verification to fail.

[0049] It is worth noting that since the update of TrustLevel in Token=H(ID+Data_hash+TrustLevel) will cause the Token value to change, in order to solve the problem of Token timeliness, the present invention adopts the following mechanism: (1) When the above-mentioned security chip 10 generates Token, it simultaneously records the timestamp T of TrustLevel; (2) When storing evidence on the blockchain, it stores the mapping relationship of {user identifier hash value, TrustLevel, timestamp T, signature information}; (3) When the AI ​​verifies the Token, it can obtain the TrustLevel corresponding to the timestamp by querying the blockchain, thereby realizing the unity of Token's "verifiability" and "timeliness".

[0050] In one embodiment of the present invention, the security chip 10 parses the unstructured raw file into key attribute fields on the local end, transmits only the encapsulated structured digital asset package, and then aggregates multiple raw detailed data into values ​​or labels that reflect statistical characteristics. Default business semantic labels are injected into the digital asset package to reduce the parsing burden on the receiving end's computing model.

[0051] It is worth noting that the aforementioned semantic tag injection includes aligning data with medical knowledge graphs, financial subject codes, or administrative attribution attributes, enabling the aforementioned AI large model interaction module 40 to directly perform decision-making reasoning based on tags.

[0052] For example, the semantic tag injection procedure executed by the aforementioned privacy protection module 30 includes precisely aligning the de-identified structured data with the default business knowledge graph: medical field alignment, mapping the original medical data to the medical knowledge graph, and converting specific physiological values ​​into standardized clinical status tags (such as disease risk level, test result classification); financial field alignment, mapping the original transaction details data to the financial account code system, and converting them into corresponding accounting account tags and statistical values; administrative field alignment, mapping the original data attributes to the administrative affiliation attribute graph, and converting them into structured geographical location or administrative division tags.

[0053] Through the aforementioned alignment mechanism, the original data is transformed into a large token with business semantics. This allows the decision-making model of the AI ​​large-scale model interaction module 40 to perform high-precision decision-making and reasoning directly based on the standardized tags without accessing the original privacy data. In this way, the amount of data transmitted can be reduced by 80% to 90% compared to the original data, achieving energy conservation and emission reduction.

[0054] In one embodiment of the present invention, the digital signature management module 20 performs signature verification (identity and permission confirmation). When a user triggers a function (such as one-click health analysis), the privacy protection module 30 will send a call request to the security chip 10 to request digital signature of the current operation command. The digital signature management module 20 will initiate multi-factor verification (such as scanning user biometrics, confirming device fingerprints and geographical location, etc.). If it is child data in the family data package, it is necessary to trigger multi-subject co-signing by both parents.

[0055] It is worth noting that the aforementioned security chip uses a private key to encrypt the instructions, generating a temporary digital signature. The aforementioned privacy protection module 30 verifies the legitimacy of the signature using a public key. Only after the aforementioned digital signature management module 20 verifies the signature, proving that the subject has been authorized and the instructions have not been tampered with, will the next step of the operation be initiated. After successful signature verification, the aforementioned privacy protection module 30 sends a de-identification instruction to the aforementioned security chip 10. The aforementioned security chip 10 completes the structured de-identification operation within the chip and transmits the de-identified large token through a secure channel to a Trusted Execution Environment (TEE) for temporary storage and subsequent processing. The aforementioned privacy protection module 30 will automatically identify the data boundaries that need to be output based on the request of the AI ​​large model (e.g., only a health label is needed, not a name).

[0056] It is worth noting that in this embodiment, the Trusted Execution Environment (TEE) serves to provide a temporary storage and running environment isolated from the operating system for the de-identified large token, preventing the de-identified data from being intercepted by malicious applications before being transmitted to the AI ​​large model; at the same time, a lightweight verification program can run within the TEE to perform secondary verification of the digital signature of the token, ensuring that the signature has not been tampered with during transmission.

[0057] It is worth noting that in this embodiment, after the digital signature management module 20 successfully verifies the signature, the privacy protection module 30 performs structured desensitization to modify the data. For example, the privacy protection module 30 directly removes personal identification information such as names and ID numbers from the original data to remove identifiers; the privacy protection module 30 aligns specific values ​​to a default knowledge graph (such as converting blood pressure values ​​into "high-risk" labels) to inject semantic tags. In a family scenario, the privacy protection module 30 retains the relationships between members (such as parent-child) but hides specific names to add family topology metadata. Finally, the privacy protection module 30 generates a proof to demonstrate to the AI ​​model that the data is authentic and conforms to the rules, but the AI ​​model still cannot see the original data.

[0058] It is worth noting that foreign legal systems (such as the EU's Electronic Identification, Authentication and Trust Services regulation (eIDAS regulation) or the US Electronic Signatures in Global and National Commerce Act (ESIGN Act) use the concept of seals less often and emphasize digital signatures more.

[0059] In one embodiment of the present invention, the aforementioned AI decision-making system 100 based on digital signature for privacy protection further includes a multilingual signature cross-domain adaptation module 50, which is used to realize the equivalent conversion and cross-border authentication of cross-domain electronic seals and digital signature standards. The aforementioned multilingual signature cross-domain adaptation module 50 can be used to extract metadata such as identity credential hash value, timestamp, and signature hierarchy from the multi-element authentication identifier (multi-certificate, multi-encryption, and multi-platform co-signature) of the aforementioned electronic seal, and convert it into an encrypted digital signature that conforms to the international PKI standard, so that the electronic seal has the same legal confirmation effect in foreign jurisdictions without a seal culture, and vice versa.

[0060] In addition, the aforementioned multilingual signature cross-domain adaptation module 50 can also repackage the above metadata into an X.509 digital certificate format that conforms to the international PKI standard, based on the target country's preset compliance policy library. The signature hierarchy is mapped to a custom field of the certificate extension field. The converted digital signature is then re-signed using an encryption algorithm recognized by the target country (such as RSA public key encryption algorithm or Elliptic Curve Digital Signature Algorithm (ECDSA)) to generate an encrypted proof document that has legal effect in the target jurisdiction.

[0061] In one embodiment of the present invention, the custom field of the above-mentioned certificate extension field can be defined using Abstract Syntax Notation dot One (ASN.1). This custom field serves as the v3 extension of the X.509 digital certificate. The extension ID can be set to an enterprise custom object identifier (OID). For example, for the Chinese region, an OID starting with 1.2.392 (assigned by the Ministry of Industry and Information Technology of the People's Republic of China) can be used, or a globally unique identifier such as 1.3.6.1.4.1.xxxxx can be applied for from the Internet Assigned Numbers Authority (IANA) can be obtained. This ensures that PKI systems in various countries can correctly parse the hierarchical relationship of the seal and the signing rules when verifying the certificate.

[0062] Please refer to Figure 2 , Figure 2 This is a flowchart of a privacy-preserving AI decision-making method based on digital signatures, as described in an embodiment of the present invention. Figure 2 As shown, the above AI decision-making method includes the following steps: S1: Receive the user's operation instructions and verify their identity and permissions through digital signature; S2: Obtain the original data of the corresponding digital asset package from the aforementioned security chip; S3: Confirm the ownership of the original data and encapsulate it into a structured large token that hides specific organizational information; S4: De-identify the large token according to the decision requirements and sign it using a digital signature to ensure the source is trustworthy; S5: Send the signed and de-identified large token to the AI ​​large model for multimodal collaborative analysis; S6: Receive and return the decision results generated by the AI ​​large model.

[0063] Example 1: Upload and Analysis of Personal Medical Test Reports

[0064] Mr. Zhang, a user, wishes to upload his hospital blood test report to an AI health platform for analysis to obtain objective advice, while ensuring that his name, ID number, and other information are not leaked. In this embodiment, Mr. Zhang clicks the "One-Click Health Analysis" button. The privacy-preserving AI decision-making system 100 based on digital signatures of this invention receives this instruction and initiates a signature request to the aforementioned digital signature management module 20. The digital signature management module 20 initiates multi-factor authentication, scanning Mr. Zhang's fingerprint and confirming the device ID. After successful verification, the aforementioned security chip 10 uses its private key to generate a temporary digital signature (step S1).

[0065] The original test report data pre-stored by Mr. Zhang is obtained from the aforementioned security chip 10 (step S2). The aforementioned security chip 10 calculates the large token primary key.

[0066] Token=H(ID+Data_hash+TrustLevel) (Formula 1)

[0067] The original data is encapsulated into an irreversible large-token digital asset package. In the formula, ID is a unique identifier, Data_hash is the hash value of the original data, and TrustLevel is the dynamic trust level (step S3).

[0068] After the digital signature management module 20 successfully verifies the signature, the privacy protection module 30 receives a security unlock signal and performs desensitization and removal of the name and ID number in the TEE environment; generalizes the age of 42 years old to an age range of [40-45); and injects the blood glucose value of 6.8 into the semantic tag clinical classification "prediabetes". Subsequently, the digital signature management module 20 uses the digital signature to digitally sign the desensitized large token (step S4).

[0069] The aforementioned AI large model interaction module 40 uploads the signed, de-identified large token to the remote AI platform (step S5).

[0070] The AI ​​platform only sees structured data and cannot obtain any original information, and returns a risk assessment report (step S6).

[0071] Example 2: Comprehensive Family Health and Education Assessment

[0072] The Wang family (with minor children) applied for government-provided comprehensive family services and was required to submit the family's health and education data. When the user clicked the "Comprehensive Family Analysis" button, the AI ​​decision-making system 100 based on digital signatures, a privacy-preserving system of this invention, triggered the aforementioned digital signature management module 20. Because the data involved children, the digital signature management module 20 forcibly initiated a multi-party co-signature process involving both parents, requiring biometric verification by both parties (step S1). The security chip 10 acquired the family data package data of the four associated members (step S2) and confirmed that they belonged to the same family data package (step S3). The privacy protection module 30 performed layered anonymization; the child's school information was completely removed; the ADHD screening results only retained the attention risk = need for observation.

[0073] It is worth noting that the privacy protection module 30 of the present invention additionally injects family topology metadata, such as three generations living together, so that the family analysis token is signed by the family's joint seal (step S4). The AI ​​big model interaction module 40 submits the de-identified big token with family topology information (step S5), and the AI ​​collaboratively outputs long-term care subsidies and after-school tutoring suggestions based on this (step S6).

[0074] Example 3: Cross-border bank account opening

[0075] Ms. Li, a Chinese citizen, opened a bank account in Singapore. While in Singapore, she selected "Cross-border Account Opening" in a mobile app. The AI ​​decision-making system 100, based on digital signatures and prior art, detected a change in geographical location, identifying Singapore as the target country. The multilingual signature module 50 invoked local compliance policies, and the digital signature management module 20 automatically activated cross-border mode and completed local identity verification using her personal seal (step S1). The security chip 10 acquired Ms. Li's identity credential data (step S2) and encapsulated it (step S3). Before signing, the system performed multilingual signature conversion, converting the Chinese personal seal to Singapore format. The desensitization process ensured that only necessary identity attributes were transmitted. The converted credential was signed with the new format seal (step S4). The AI ​​large-scale model interaction module 40 submitted the credential to the Singapore bank system (step S5). The data volume was small, aligning with energy conservation and emission reduction principles. After successful bank verification, the bank account was opened (step S6).

[0076] Example 4: Application of Zero-Knowledge Proofs in Credit Approval

[0077] When a user applies for a loan, they need to prove to the bank's AI that their credit score is higher than 700, but they do not wish to disclose the specific score. After verifying the user's identity, the privacy protection module 30 prepares to generate the proof (step S1). The security chip 10 obtains the user's credit score (e.g., 720) (step S2) and encapsulates it (step S3). The privacy protection module 30 generates a zero-knowledge proof P, which is itself a special type of anonymized output. Subsequently, the digital signature management module 20 signs the zero-knowledge proof P and the related big token (step S4). The AI ​​big model interaction module 40 submits the zero-knowledge proof P to the bank's AI (step S5). After the AI ​​big model verifies the validity of the zero-knowledge proof P, it directly approves the loan (step S6). Throughout the entire process, the user's precise credit score is not exposed, perfectly demonstrating the core value of data being usable but invisible.

[0078] The embodiments of the present invention described above can be implemented in various hardware, software codes, or combinations thereof. For example, embodiments of the present invention can also be program code executing the above methods in a Digital Signal Processor (DSP). The present invention can also relate to various functions executed by a computer processor, digital signal processor, microprocessor, or Field Programmable Gate Array (FPGA). The processor described above can be configured to perform specific tasks according to the present invention, which are accomplished by executing machine-readable software code or firmware code defining the specific methods disclosed in the present invention. The software code or firmware code can be developed into different programming languages ​​and different formats or forms. The software code can also be compiled for different target platforms. However, the different code styles, types, and languages ​​of the software code performing tasks according to the present invention and other types of configuration code do not depart from the spirit and scope of the present invention.

[0079] Therefore, those skilled in the art will recognize that although embodiments of the present invention have been shown and described in detail herein, many other variations or modifications conforming to the principles of the present invention can be directly determined or derived from the disclosure of the present invention without departing from the spirit and scope of the invention. Therefore, the scope of the present invention should be understood and recognized as covering all such other variations or modifications.

Claims

1. A privacy-preserving AI decision-making system based on digital signatures, comprising: A security chip with a built-in cryptographic hash function H is used to calculate based on the user's unique identifier ID, the original data hash value Data_hash, and the dynamic trust level TrustLevel. Token=H(ID+Data_hash+TrustLevel) Generate an irreversible large token digital asset package; A digital signature management module is used to generate a multi-level digital signature system, bind digital signatures with multi-dimensional organizational information, and digitally sign large token digital asset packages. A privacy protection module is used to call the digital signature management module for signature verification and trigger the de-identification procedure to ensure that the original information does not leave the domain; as well as An AI large-scale model interaction module is used to receive signed and anonymized large-scale token digital asset packages, generate decision results using AI analysis capabilities, and does not touch the original sensitive privacy data during the processing.

2. The privacy-preserving AI decision-making system based on digital signature according to claim 1, characterized in that, It also includes a multilingual signature cross-domain adaptation module, used to achieve equivalent conversion between cross-domain electronic seals and digital signature standards and cross-border authentication. The multilingual signature cross-domain adaptation module is used for: The multi-element authentication identifier of the source digital signature is parsed, and the metadata is extracted. The metadata includes the identity credential hash value, timestamp, and signature hierarchy. Based on the target country's pre-defined compliance policy database, the metadata is repackaged into an X.509 digital certificate format conforming to international PKI standards, wherein the signature hierarchy is mapped to custom fields in the certificate extension field; and The converted digital signature is then re-signed using an encryption algorithm recognized by the target country, generating an encrypted certificate that is legally valid in the target jurisdiction.

3. The privacy-preserving AI decision-making system based on digital signature according to claim 1, characterized in that: The large token digital asset package has a dynamic trust level, which is updated according to the user's credit behavior. After the updated trust level is digitally signed, it is packaged and uploaded to the blockchain network for evidence storage. The evidence storage content includes the user identifier hash value, the new trust level, the timestamp, and the signature information.

4. The privacy-preserving AI decision-making system based on digital signature according to claim 2, characterized in that: The large token digital asset package has a dynamic trust level, which is updated according to the user's credit behavior. After the updated trust level is digitally signed, it is packaged and uploaded to the blockchain network for evidence storage. The evidence storage content includes the user identifier hash value, the new trust level, the timestamp, and the signature information.

5. The privacy-preserving AI decision-making system based on digital signature according to claim 1, characterized in that: The digital signature management module performs signature verification. When a user triggers a function, the privacy protection module sends a call request to the security chip, requesting a digital signature for the current operation instruction. The digital signature management module initiates multi-factor authentication, and the security chip uses its private key to encrypt the call request instruction, generating a temporary digital signature.

6. The privacy-preserving AI decision-making system based on digital signature according to claim 5, characterized in that: The privacy protection module verifies the legitimacy of the temporary digital signature using a public key. After successful signature verification, the privacy protection module sends a de-identification instruction to the security chip. The security chip performs structured de-identification within the chip and transmits the de-identified large token to a trusted execution environment for temporary storage and subsequent processing via a secure channel.

7. The privacy-preserving AI decision-making system based on digital signatures according to any one of claims 1 to 6, characterized in that: The privacy protection module parses unstructured raw files of multi-dimensional data into key attribute fields and transmits only the encapsulated structured digital asset package; it aggregates multiple raw data into values ​​or labels that reflect statistical characteristics; and it injects default business semantic labels into the digital asset package to reduce the parsing burden on the receiving end's computing model.

8. The privacy-preserving AI decision-making system based on digital signature according to claim 7, characterized in that, The semantic tag injection procedure executed by the privacy protection module includes: precisely aligning the de-identified structured data with the default business knowledge graph; wherein, the medical domain alignment maps the original medical data to the medical knowledge graph, converting specific physiological values ​​into standardized clinical status tags; the financial domain alignment maps the original bank statement data to the financial account code system, converting it into corresponding accounting account tags and statistical values; and the administrative domain alignment maps the original data attributes to the administrative affiliation attribute graph, converting it into structured geographical location or administrative division tags.

9. A privacy-preserving AI decision-making method based on digital signatures, comprising the following steps: S1: Receive user operation instructions and verify their identity and permissions through signature; S2: Obtain the original data of the corresponding digital asset package from the security chip; S3: Confirm ownership of the original data and encapsulate it into a structured large token that hides specific organizational information; S4: Structure and desensitize large tokens according to decision-making needs, and use digital signatures to sign them to ensure the source is trustworthy; S5: Send the desensitized token after signing to the AI ​​big model for multimodal collaborative analysis; S6: Receive and return the decision results generated by the large AI model.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, wherein, when the computer program is run by a processor, it controls the device where the storage medium is located to execute the privacy-preserving AI decision-making method based on digital signature as described in claim 9.

Images