A trusted model context interaction method and apparatus

Abstract

The present application relates to the technical field of artificial intelligence, and provides a trusted model context interaction method and device, which comprises the following steps: determining trusted storage resource content and / or execution tools required for executing a target task, sending calling parameters representing the resource content and / or the execution tools to a server, so that the server performs identity verification and calling confidence determination on the calling parameters; when it is determined that the identity verification is passed and the calling confidence meets the standard, obtaining the resource content and / or the execution tools from a block chain according to the calling parameters, and executing the target task according to the obtained resource content and / or the execution tools to obtain an execution result; receiving the execution result, performing identity verification and result confidence determination on the execution result, and performing a task delivery operation on the execution result when it is determined that the identity verification is passed and the execution confidence meets the standard. The present application can resist attack risks such as malicious tool forgery, parameter tampering and result tampering, and provides a solid guarantee for the safety of intelligent agents.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a reliable model context interaction method and apparatus. Background Technology

[0002] With the development of Large Language Models (LLMs) and their Model Context Protocols (MCPs), LLMs have been rapidly applied in scenarios such as intelligent agents, automated decision-making, and knowledge services. Intelligent agents often need to call external tools and resources through Model Context Protocols (MCPs). Currently, MCPs have demonstrated great potential in scenarios such as manufacturing quality analysis, statistical interface calls, and private domain knowledge base updates. However, there are still significant shortcomings in ensuring trustworthiness when calling external tools and resources through Model Context Protocols (MCPs), leading to potential vulnerabilities in the system when facing audits, result traceability, and accountability. Summary of the Invention

[0003] To address the problems existing in the prior art, the present invention provides a reliable model context interaction method and apparatus.

[0004] This invention provides a reliable model context interaction method, comprising:

[0005] The intelligent agent sends the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration credibility, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration credibility.

[0006] The intelligent agent determines the resources and / or execution tools required to execute the target task, and sends the calling parameters representing the resources and / or execution tools to the server so that the server can perform identity verification and call confidence judgment on the calling parameters. When it is determined that the identity verification is successful and the call confidence meets the standard, the intelligent agent obtains the resources and / or execution tools from the blockchain according to the calling parameters, and executes the target task according to the obtained resources and / or execution tools to obtain the execution result.

[0007] The intelligent agent receives the execution result, performs identity verification and result confidence judgment on the execution result, and performs task delivery operation on the execution result when it is determined that the identity verification is successful and the execution credibility meets the standard.

[0008] According to a trusted model context interaction method provided by the present invention, the method further includes:

[0009] Before the intelligent agent sends the calling parameters representing the resource content and / or execution tool to the server, the calling parameters are signed.

[0010] According to a trusted model context interaction method provided by the present invention, the step of performing identity verification and result confidence judgment on the execution result includes:

[0011] The execution result is signed and its identity is verified. The execution result is signed by the server and then fed back to the intelligent agent.

[0012] The data structure of the execution result is verified item by item against the data structure of the data output mode registered by the agent, and an execution credibility characterization of the execution result is generated.

[0013] According to a trusted model context interaction method provided by the present invention, the method further includes:

[0014] Once the signature and identity validity verification is passed and the execution credibility is within the sampling inspection score range, the task delivery operation is performed on the execution result, and the sampling inspection rate of the execution result is increased.

[0015] When it is determined that the signature and identity validity verification is successful and the execution credibility is within the range of the rejection delivery score, the execution result is rejected for delivery.

[0016] According to a trusted model context interaction method provided by the present invention, the method further includes:

[0017] Receive the registration credibility of the intelligent agent's registered resource content and / or execution tools;

[0018] The overall credibility is determined based on the registration credibility, invocation credibility, and execution credibility.

[0019] Based on the overall credibility, the servers corresponding to the overall credibility are classified into different levels.

[0020] This invention also provides a reliable model context interaction method, comprising:

[0021] The server receives the description content of the execution tool and resource content to be registered from the intelligent agent. After performing hash calculation and signature processing on the description content of the execution tool and resource content, it sends it to the blockchain so that the blockchain can sign and verify the legality of the description content of the execution tool and resource content, compare it with the key fields of the content, obtain the registration credibility, and make a judgment on the storage of the description content of the execution tool and resource content based on the registration credibility.

[0022] The server receives the call parameters sent by the intelligent agent, performs identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call credibility meets the standard, it retrieves resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result;

[0023] The server sends the execution result to the agent, so that the agent can perform identity verification and result confidence judgment on the execution result. When it is determined that the identity verification is successful and the execution credibility meets the standard, the agent performs task delivery operation on the execution result.

[0024] According to a trusted model context interaction method provided by the present invention, the step of performing identity verification and call confidence judgment on the call parameters includes:

[0025] The call parameters are verified by signature, identity validity, and timeliness.

[0026] The data structure of the calling parameters and the data structure of the agent registration are compared by field comparison to generate a calling credibility characterizing the calling parameters.

[0027] The present invention also provides a trusted model context interaction device, comprising:

[0028] The first registration trust module is used to send the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration trust, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration trust.

[0029] The first call trust module is used to determine the resource content and / or execution tools required to execute the target task, and send call parameters representing the resource content and / or execution tools to the server so that the server can perform identity verification and call trust judgment on the call parameters. When it is determined that the identity verification is successful and the call trust is up to standard, the server retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result.

[0030] The first execution trust module is used to receive the execution result, perform identity verification and result confidence judgment on the execution result, and perform task delivery operation on the execution result when it is determined that the identity verification is successful and the execution trust is up to standard.

[0031] The present invention also provides a trusted model context interaction device, comprising:

[0032] The second registration trust module is used to receive the description content of the execution tool and the resource content to be registered sent by the intelligent agent, perform hash calculation and signature processing on the description content of the execution tool and the resource content and send them to the blockchain, so that the blockchain can sign and verify the legality of the description content of the execution tool and the resource content, compare them using the key fields of the content, obtain the registration trust, and make a judgment on the storage of the description content of the execution tool and the resource content based on the registration trust.

[0033] The second trusted module is used to receive the call parameters sent by the intelligent agent, perform identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it obtains resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the obtained resource content and / or execution tools to obtain the execution result.

[0034] The second execution trust module is used to send the execution result to the intelligent agent so that the intelligent agent can perform identity verification and result confidence judgment on the execution result. When it is determined that the identity verification is successful and the execution trust is up to standard, the intelligent agent performs task delivery operation on the execution result.

[0035] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement any of the trusted model context interaction methods described above.

[0036] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the trusted model context interaction methods described above.

[0037] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the trusted model context interaction methods described above.

[0038] This invention provides a trusted model context interaction method and apparatus. By making the large model context protocol trustworthy in the tool registration, invocation and result stages, a trusted closed-loop system covering the entire link is established. This system can resist the attack risks such as malicious tool forgery, parameter tampering and result tampering, and provides a complete and verifiable chain of evidence in compliance audit and accountability scenarios. Thus, it provides a solid guarantee for the secure application of large model agents in cross-domain and complex environments. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0040] Figure 1 This is a flowchart illustrating the trusted model context interaction method provided by the present invention. Figure 1 .

[0041] Figure 2 This is a diagram illustrating the implementation process of the trusted model context interaction method provided by this invention.

[0042] Figure 3 This is a flowchart illustrating the trusted model context interaction method provided by the present invention. Figure 2 .

[0043] Figure 4 This is a schematic diagram of the structure of the trusted model context interaction device provided by the present invention. Figure 1 .

[0044] Figure 5 This is a schematic diagram of the structure of the trusted model context interaction device provided by the present invention. Figure 2 .

[0045] Figure 6 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0047] Figure 1 The diagram illustrates a flowchart of a trusted model context interaction method provided by this invention. (See attached diagram.) Figure 1 This method uses the intelligent agent as the execution subject and includes the following steps:

[0048] Step 11: The intelligent agent sends the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration credibility, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration credibility.

[0049] Step 12: The intelligent agent determines the resources and / or execution tools required to execute the target task, and sends the calling parameters representing the resources and / or execution tools to the server so that the server can verify the identity and make a call confidence judgment on the calling parameters. When it is determined that the identity verification is successful and the call confidence meets the standard, the intelligent agent obtains the resources and / or execution tools from the blockchain according to the calling parameters, and executes the target task according to the obtained resources and / or execution tools to obtain the execution result.

[0050] Step 13: The intelligent agent receives the execution result, performs identity verification and result confidence judgment on the execution result, and delivers the task to the execution result when the identity verification is successful and the execution credibility meets the standard.

[0051] Regarding steps 11-13, it should be noted that with the development of Large Language Models (LLMs) and their Model Context Protocols (MCPs), LLMs have been rapidly applied in scenarios such as intelligent agents, automated decision-making, and knowledge services. Intelligent agents often need to call external tools and resources through Model Context Protocols (MCPs), and these resources are used by the intelligent agent to perform tasks. The content types of these resources include databases, files, APIs, and other computational models. Currently, MCPs have demonstrated great potential in scenarios such as manufacturing quality analysis, statistical interface calls, and private domain knowledge base updates. However, there are still significant shortcomings in ensuring trustworthiness when calling external tools and resources through Model Context Protocols (MCPs), leading to potential vulnerabilities in the system when facing audits, result traceability, and accountability.

[0052] In response, this invention focuses on the interaction between large-model-driven intelligent agents and servers / resources, requiring systematic improvements in trustworthiness, traceability, and security. Therefore, this invention proposes a closed-loop design for the entire chain of trusted registration, trusted invocation, and trusted execution, ensuring the authenticity, integrity, and traceability of resources and execution tools throughout the entire process from entering the system to generating results.

[0053] In the trusted registration phase, this invention aims to store the execution tools and resources to be registered on the blockchain and perform independent identity verification, making the origin of the tools and resources transparent and difficult to tamper with. Therefore, the intelligent agent in this invention sends the description of the execution tools and resource content to be registered to the server. The execution tools are like the "hands and feet" extended from the intelligent agent. When the intelligent agent needs to do something it cannot do directly (such as checking the weather, executing code, or manipulating files), it can call the corresponding "tools" to do it for it.

[0054] The server performs hash calculations and signature processing on the description of the execution tool and the resource content before sending them to the blockchain.

[0055] The blockchain signs and verifies the legality of the description and resource content of the execution tool, compares them using key content fields to obtain the registration credibility, and makes a judgment on the storage of the description and resource content of the execution tool based on the registration credibility (if the registration credibility meets the expectations, the execution tool and resources can be stored on the blockchain).

[0056] During the trusted registration phase, this invention not only employs a blockchain notarization mechanism, but also hashes and signs the tool description and registration information before writing it onto the chain to ensure the authenticity and immutability of the registration information.

[0057] For signature and legitimacy verification, an access control verification and field-level commitment mechanism are implemented to quickly verify the legitimacy of server-side signatures, certificates, and timestamps; if verification fails, the application is rejected outright. Subsequently, key fields such as the name, input schema, output schema, and description of the execution tool or resource are checked and compared one by one, avoiding the coarseness of relying solely on overall hash comparison. Finally, based on the verification and comparison results and an aggregate scoring mechanism, a registration trustworthiness score is output to determine whether the execution tool or resource is included in the trusted database, achieving refined trust assurance for tool registration.

[0058] In this invention, the intelligent agent gradually advances and completes complex target tasks through repeated thinking (reasoning), action (tool invocation), and observation (feedback acquisition). Therefore, when the intelligent agent performs a task, it needs to invoke corresponding execution tools and / or resources to complete the task.

[0059] For example, if someone asks, "Please check the weather in Beijing today," the intelligent agent needs to call a weather query tool (such as connecting to a weather application software as the execution tool) to obtain the weather information for Beijing.

[0060] In the trusted invocation phase, this invention aims to prevent invocation requests from being forged or tampered with, addressing the risks of unauthorized invocations and request impersonation. Therefore, the intelligent agent in this invention needs to perform reasoning analysis on the target task to obtain the resource content and / or execution tools required to execute the target task, and then generate invocation parameters representing the resource content and / or execution tools.

[0061] The call parameters are sent to the server. The server verifies the identity and trustworthiness of the call parameters. If the identity verification is successful and the call trustworthiness meets the standard, the server retrieves the resource content and / or execution tool from the blockchain based on the call parameters. This enables the acquisition of the corresponding resource content and / or execution tool under trusted call parameters.

[0062] The server then executes the target task based on the acquired resources and / or execution tools, obtains the execution result, and feeds the execution result back to the agent.

[0063] During the trusted invocation phase, this invention introduces a two-way authentication and parameter signing mechanism. The invocation parameters are signed using a three-phase trusted context signing (3TSig) mechanism, followed by access control verification (signature verification, identity legitimacy, and timeliness checks). Simultaneously, a field-level comparison mechanism is used to compare the invocation parameters of legitimate individuals, confirming the consistency between the invocation content and the registered commitment. Once the comparison result is successful, the invocation scoring mechanism outputs a call trustworthiness score, which, combined with ACL access control, quota restrictions, and schema verification mechanisms, determines whether the invocation should be executed directly or rejected. This ensures the authenticity, integrity, and legitimacy of the invocation process, avoiding the risks of forged calls and parameter tampering.

[0064] In the trusted execution phase, the execution result is directly returned to the client (or agent), aiming to ensure that the result is not tampered with during transmission or processing, and that the client can verify the reliability of its source, thus avoiding the risk of result forgery or tampering.

[0065] During the trusted execution phase, the agent (or client) receives the execution result, performs identity verification and result confidence judgment on the execution result, and delivers the task to the execution result when it is determined that the identity verification is successful and the execution trustworthiness meets the standard.

[0066] This invention requires the server to perform hash calculations and generate signatures on the result data after generating the execution result, and, if necessary, write the result digest to the blockchain for evidence storage. Upon receiving the result, the intelligent agent first performs access control verification to complete the signature and identity legitimacy verification. Then, it performs item-by-item verification of the data structure of the execution result and the output schema (output_schema) during registration to generate an execution credibility score. If the execution credibility score is high, the result is delivered directly; if it is in the middle range, it is delivered but the sampling rate is increased; if it is too low, it is rejected and an audit is triggered. Through this mechanism, the result process not only possesses authenticity and immutability but also provides complete on-chain evidence in the event of a dispute.

[0067] Furthermore, this invention proposes a server-side credibility aggregation and scoring mechanism. By weighted summarizing registration credibility, call credibility, and execution credibility, an overall server credibility score is formed and divided into three levels—high, medium, and low—based on thresholds, corresponding to priority calls, normal calls and high-sampling inspections, and downgraded calls and manual audits, respectively. This enables the intelligent agent to dynamically manage the server-side credibility.

[0068] Furthermore, this invention introduces a three-stage trusted context signature (3TSig) mechanism into the digital signature stage. It defaults to the Ed25519 algorithm and can be extended to BLS aggregate signature or threshold signature in high-value or high-risk scenarios, thereby enhancing the system's non-repudiation and robustness while maintaining efficiency. Under this signature mechanism, a context field is introduced into the message body of each stage, and signature domains are separated to avoid replay attacks across stages or sessions. In high-risk scenarios, threshold signature or aggregate signature is supported to achieve multi-party endorsement and higher security levels. This not only enhances the reliability of registration, invocation, and result verification but also provides a solid foundation for future multi-agent collaboration, cross-domain invocation, and compliance auditing.

[0069] The trusted model context interaction method provided by this invention establishes a trusted closed-loop system covering the entire chain by making the large model context protocol trustworthy in the tool registration, invocation and result stages. This system can resist the attack risks such as malicious tool forgery, parameter tampering and result tampering, and provides a complete and verifiable chain of evidence in compliance audit and accountability scenarios. Thus, it provides a solid guarantee for the secure application of large model agents in cross-domain and complex environments.

[0070] The implementation process of the reliable model context interaction method proposed in this invention can be combined with... Figure 2The overall framework diagram shown below illustrates this. The overall architecture comprises four core components: the agent / user side, the client side, the server side, and the blockchain evidence storage system. These components play a role in the three stages of trusted registration, trusted invocation, and trusted execution, respectively, forming a full-chain trusted assurance system covering the entire lifecycle of the tool.

[0071] During the trusted registration phase, the request trigger unit on the Agent / User side initiates the tool invocation intent. The request reaches the tool registration verification module on the Client side, which performs two key functions: first, it checks whether the target tool already exists in the local trusted tool library; second, if the tool has not been registered or the certificate information has expired, it triggers a new round of registration process.

[0072] In the registration process, the server-side tool description submission module is responsible for generating the core description information (desc) of the tool, including metadata such as tool name, input schema, output schema, and description. The server first performs normalization processing on the overall description information and calculates its description hash:

[0073]

[0074] in, This represents the digest value obtained by hashing the core description information, used to generate a unique identifier for the tool. For hash algorithms, secure hash algorithms such as SHA-256 and SM3 can be used.

[0075] The server then uses its own private key to generate a signature:

[0076]

[0077] This indicates that the server uses its private key sk. server The value of the hash result signature is used to verify the authenticity of the registration source.

[0078] Forming a registration information package:

[0079]

[0080] This represents the server's public key, used to verify the validity of its signature. The timestamp represents the registration or invocation of an event, used to prevent replay attacks and ensure the consistency of the timing of operations.

[0081] This information first enters the registration and access control verification unit. This unit verifies the consistency between the signature and the public key, ensuring that the information submitted by the server possesses basic authenticity and non-repudiation. If the access control verification fails, the registration request is immediately rejected to prevent forgery tools from entering the system.

[0082] After passing the access control verification, the server submits the registration information to the blockchain evidence storage system for registration and evidence storage. Module The system is responsible for writing the tool description hash and signature information onto the blockchain to ensure the authenticity and immutability of the registration data. Simultaneously, the client-side field-level commitment comparison unit calculates the hash of each key field in the tool description.

[0083]

[0084] in, This refers to each field in the tool description. Perform hashing independently to prevent field content from being tampered with. Here, i is the field index. Indicates the field The input is fed into a hash algorithm to calculate its digest.

[0085] The calculated hash value is compared with the corresponding commitment value stored in the blockchain. Unlike traditional Boolean-based judgment mechanisms, this invention introduces a fractionalized credibility evaluation at this stage: if all fields are consistent, a credibility score is registered. A perfect score is awarded; if some fields differ, a portion of the score is allocated according to the field's weight; if key fields are inconsistent, registration fails immediately. Furthermore, the signature Sig(server) generated by the server at this stage also serves as the first-stage signature in the three-stage trusted context signature (3TSig) and is bound to a context identifier (CtxID), providing the foundation for subsequent calls and result signature concatenation. Finally, the client-side tool registration verification module integrates the results of access control verification and field-level commitment comparison to generate a registration trust score. When the score exceeds the threshold, the tool is included in the trusted tool library for use in subsequent calls; if it falls below the threshold, registration is rejected, and the exception information is recorded in the audit log.

[0086] Through the above mechanisms, the registration process not only verifies the authenticity and integrity of the tool's identity, but also avoids malicious descriptions and poisoning through access control verification and field-level comparison; at the same time, the 3TSig scoring mechanism provides a dynamic credibility reference for subsequent calling processes, thereby making the tool registration trustworthy.

[0087] During the trusted invocation phase, the trusted guarantee of the invocation request and execution process is achieved through identity authentication, invocation signature, parameter verification and invocation evidence storage mechanisms.

[0088] First, the MCP client and MCP server establish a secure communication channel by calling the handshake and authentication modules. Both parties use certificates or on-chain identities for two-way authentication and generate temporary session keys based on a key exchange algorithm, thereby ensuring the security and non-repudiation of the communication process.

[0089] After the communication channel is established, the MCP client will perform a hash calculation on the call parameters:

[0090]

[0091] Then, a signature is generated using one's own private key:

[0092]

[0093] It indicates that the client uses its private key. For call parameters The hash value is used to sign the request, which is used to prove the authenticity and integrity of the call request.

[0094] The final call request packet is formed as follows:

[0095]

[0096] The request first enters the access control verification unit, which is responsible for verifying the Sig (client) and checking whether the call request remains valid under contextual conditions such as timestamps and session keys. If the access control verification fails, the call request is immediately rejected, and the exception is written to the audit log to prevent forged or expired requests from entering the execution process. After passing the access control verification, the request enters the call field comparison unit. This unit performs consistency verification on each field of the request parameters against the expected schema to ensure that the type, format, and range of the input parameters comply with system specifications. Simultaneously, it combines ACL (Access Control List) permission control and quota restriction mechanisms to impose dual restrictions on the caller's identity and resource usage, preventing unauthorized calls, resource abuse, or malicious calls. After completing the field-level verification, the system generates a credibility score for this call. The scoring criteria include parameter completeness, schema compliance, and ACL check results. Unlike traditional Boolean-based judgment mechanisms, this invention employs a fractional evaluation model: the highest score is given for complete consistency, partial scores are awarded for partial deviations, and calls are directly rejected for serious discrepancies.

[0097] Meanwhile, the signature generated by the MCP client at this stage This signature will serve as the second-stage signature in the three-stage trusted context signature (3TSig) process, bound to the same context identifier (CtxID) as the signature from the registration stage. This not only ensures context consistency between the invocation and registration stages but also prevents cross-stage forgery and replay attacks. Only after the invocation request passes the above verification will the MCP server enter the tool execution module, execute the target tool logic, and generate the invocation result. During invocation execution, the server-side audit log module will record the invocation summary, execution status, and resource consumption in real time, and can selectively submit the invocation summary to the invocation notarization module of the blockchain notarization system, thus forming an immutable invocation record.

[0098] Through the above mechanisms, the invocation process not only verifies the integrity of the request and the authenticity of the identity, but also prevents malicious invocation and parameter tampering through access control verification and field comparison. At the same time, the score-based scoring and 3TSig mechanism provide dynamic credibility assessment for the invocation process, enabling the system to accumulate and quantify the overall credibility of the server in multiple interactions.

[0099] During the trusted execution phase, the authenticity, immutability, and traceability of the output results are ensured through result signing, result verification, and result storage mechanisms.

[0100] After the MCP server finishes executing the tool, it will perform a hash calculation on the result data:

[0101]

[0102] Then, a signature is generated using one's own private key:

[0103]

[0104] The above formula indicates that the server signs the hash value of the output result 'out', which is used for integrity verification and traceability proof of the call result.

[0105] Result return packet:

[0106]

[0107] The returned result packet first enters the result access control verification unit. This unit verifies Sig(server), checking if it matches the server's public key, and confirms that the signature context identifier (CtxID) is consistent with the previous two stages. If access control verification fails, the result is immediately rejected, and the event is written to the audit log.

[0108] After passing the access control verification, the result return packet enters the result field comparison unit. This unit mainly performs consistency checks on the result summary, output schema (output_schema), or, if necessary, key result fields to ensure that the returned result has not been tampered with or forged during transmission. Similar to the registration and invocation phases, this step also employs a score-based evaluation mechanism to generate an execution reliability score. The highest score is awarded if all fields are completely identical; a slight discrepancy will result in a lower score; and results will be rejected if key fields are inconsistent.

[0109] After verification and scoring are completed, the result return packet enters the result signing and storage module. This module will, as needed,... The submission is made to the blockchain evidence storage system, forming an immutable evidence record at the result level, ensuring a strong chain of evidence for external audits or dispute arbitration. Simultaneously, the MCP client uses the Server's public key to verify the signature and recalculates the result hash, comparing it with the returned value to confirm the integrity and authenticity of the result.

[0110] In addition, the signature in the result phase This signature will serve as the third stage signature in the three-stage trusted context signature (3TSig) process, forming a complete context signature chain together with the registration and invocation stage signatures. The verifier can trace and verify the consistency of the three-stage signatures using the CtxID, ensuring that the result is not only trustworthy in a single stage but also possesses complete verifiability across cross-stage interactions. Finally, the MCP client's result verification module delivers the result to the Agent / User's result receiver if both signature verification and field comparison pass; if verification fails, the result is rejected, and alarms and audit logs are triggered.

[0111] Through the above mechanisms, the trusted execution phase not only ensures the authenticity and integrity of the results, but also enhances the ability to resist tampering and forgery through access control verification and field-level commitment mechanisms. At the same time, combined with the score-based scoring and 3TSig signature chain, it ensures that the trustworthiness of the execution is quantifiable and traceable, and forms a trusted closed loop with the registration and invocation links.

[0112] In this invention, the auditing and traceability mechanism serves as the core link connecting the three stages of trusted registration, trusted invocation, and trusted execution, ensuring the system's verifiability and traceability throughout its entire lifecycle. Its implementation path includes three parts: client-side auditing, server-side log management, and blockchain-based evidence storage.

[0113] First, on the agent (or client) side, the auditing and traceability module generates a summary record for each tool registration, invocation, and result return process, including message hash, signature verification status, trust score, and timestamp. This module not only supports local log management but can also selectively write key summaries to the blockchain, thereby ensuring the immutability and persistent verifiability of the data.

[0114] Secondly, on the server side, the audit log module is responsible for recording information throughout the entire process of the call execution, such as tool execution status, resource consumption, abnormal events, and signature generation process. These logs can be cross-referenced with client records and solidified through the blockchain's call and result notarization modules, ensuring a unified reference basis in case of disputes or security audits.

[0115] Furthermore, on the blockchain side, the verification support module provides a unified interface for external auditing and arbitration. Any third party that needs verification can call this module to check the registration evidence, the call evidence, and the result evidence, thereby reconstructing a complete chain of interactive evidence.

[0116] Based on the above, a registration credibility score is calculated. Calling a reliable score Execute credible scoring Perform weighted calculation:

[0117]

[0118] in, These represent the weights of the three stages in different application scenarios, and can be flexibly adjusted according to business needs. This formula allows for the generation of an overall trustworthiness score for the server. .

[0119] Based on the aggregated scoring results, the system classifies the server's trustworthiness into three levels:

[0120] 1. High credibility: If the value exceeds the upper threshold, the call request enters the fast track, reducing the sampling rate and prioritizing the scheduling of this server.

[0121] 2. Moderate credibility: When the threshold is between the upper and lower limits, the call request is executed according to the normal process, while the sampling rate is increased for additional verification.

[0122] 3. Low credibility: If the request falls below the lower threshold, the call will be downgraded and manual auditing will be triggered if necessary to ensure that abnormal behavior is intercepted in a timely manner.

[0123] Figure 3The diagram illustrates a flowchart of a trusted model context interaction method provided by this invention. (See attached diagram.) Figure 3 This method takes the server side as the execution subject and includes the following steps:

[0124] Step 31: The server receives the description content of the execution tool and resource content to be registered from the intelligent agent. After performing hash calculation and signature processing on the description content and resource content, it sends them to the blockchain so that the blockchain can sign and verify the legality of the description content and resource content, compare them using key content fields, obtain the registration credibility, and determine whether to store the description content and resource content in the database based on the registration credibility.

[0125] Step 32: The server receives the call parameters sent by the agent, performs identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it obtains the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the obtained resource content and / or execution tools to obtain the execution result; the call parameters are the call parameters of the resource content and / or execution tools required for executing the target task as determined by the agent.

[0126] Step 33: The server sends the execution result to the agent so that the agent can verify the identity and make a confidence judgment on the execution result. When the identity verification is successful and the execution credibility meets the standard, the agent performs the task delivery operation on the execution result.

[0127] A further step in the above method involves verifying the identity of the calling parameters and determining the calling confidence, including:

[0128] Perform signature verification, identity validity and timeliness checks on the call parameters;

[0129] The data structure of the calling parameters and the data structure of the agent registration are compared by field comparison to generate a calling credibility characterization of the calling parameters.

[0130] Since the method in this embodiment of the invention is based on the same principle as the method in the above embodiments, more detailed explanations will not be repeated here.

[0131] The trusted model context interaction device provided by the present invention will be described below. The trusted model context interaction device described below can be referred to in correspondence with the trusted model context interaction method described above.

[0132] Figure 4 A schematic diagram of the structure of a trusted model context interaction device provided by the present invention is shown. See also... Figure 4 The device includes a first registration trusted module 41, a first invocation trusted module 42, and a first execution trusted module 43, wherein:

[0133] The first registration trust module is used to send the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration trust, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration trust.

[0134] The first call trust module is used to determine the resource content and / or execution tools required to execute the target task, and send call parameters representing the resource content and / or execution tools to the server so that the server can perform identity verification and call trust judgment on the call parameters. When it is determined that the identity verification is successful and the call trust is up to standard, the server retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result.

[0135] The first execution trust module is used to receive the execution result, perform identity verification and result confidence judgment on the execution result, and perform task delivery operation on the execution result when it is determined that the identity verification is successful and the execution trust is up to standard.

[0136] Since the apparatus of this embodiment is based on the same principle as the method of the above embodiment, more detailed explanations will not be repeated here.

[0137] It should be noted that, in the embodiments of the present invention, the relevant functional modules can be implemented by a hardware processor.

[0138] The trusted model context interaction device provided by this invention establishes a trusted closed-loop system covering the entire chain by making the large model context protocol trustworthy in the tool registration, invocation and result stages. This system can resist the attack risks such as malicious tool forgery, parameter tampering and result tampering, and provides a complete and verifiable chain of evidence in compliance audit and accountability scenarios. Thus, it provides a solid guarantee for the secure application of large model agents in cross-domain and complex environments.

[0139] Figure 5 A schematic diagram of the structure of a trusted model context interaction device provided by the present invention is shown. See also... Figure 5 The device includes a second registration trusted module 51, a second invocation trusted module 52, and a second execution trusted module 53, wherein:

[0140] The second registration trust module is used to receive the description content of the execution tool and resource content to be registered sent by the intelligent agent, perform hash calculation and signature processing on the description content of the execution tool and resource content and send it to the blockchain so that the blockchain can sign and verify the legality of the description content of the execution tool and resource content, compare it with the key fields of the content to obtain the registration trust level, and make a judgment on the storage of the description content of the execution tool and resource content based on the registration trust level.

[0141] The second trusted module is used to receive the call parameters sent by the intelligent agent, perform identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it retrieves resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result.

[0142] The second execution trust module is used to send the execution result to the agent so that the agent can verify the identity of the execution result and make a confidence judgment on the result. When it is determined that the identity verification is successful and the execution trust is up to standard, the agent performs the task delivery operation on the execution result.

[0143] Since the apparatus of this embodiment is based on the same principle as the method of the above embodiment, more detailed explanations will not be repeated here.

[0144] It should be noted that, in the embodiments of the present invention, the relevant functional modules can be implemented by a hardware processor.

[0145] The trusted model context interaction device provided by this invention establishes a trusted closed-loop system covering the entire chain by making the large model context protocol trustworthy in the tool registration, invocation and result stages. This system can resist the attack risks such as malicious tool forgery, parameter tampering and result tampering, and provides a complete and verifiable chain of evidence in compliance audit and accountability scenarios. Thus, it provides a solid guarantee for the secure application of large model agents in cross-domain and complex environments.

[0146] Figure 6 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 6 As shown, the electronic device may include: a processor 61, a communication interface 62, a memory 63, and a communication bus 64, wherein the processor 61, the communication interface 62, and the memory 63 communicate with each other via the communication bus 64. The processor 61 can invoke logical instructions in the memory 63 to execute a trusted model context interaction method, which includes:

[0147] Send the description of the execution tool and the resource content to be registered to the server. The server will then perform hash calculation and signature processing on the description of the execution tool and the resource content before sending it to the blockchain. The blockchain will then sign and verify the legality of the description of the execution tool and the resource content, and compare it using key content fields to obtain the registration credibility. Based on the registration credibility, the description of the execution tool and the resource content will be entered into the database.

[0148] The system determines the resources and / or execution tools required to perform the target task, sends call parameters representing the resources and / or execution tools to the server, so that the server can perform identity verification and call confidence judgment on the call parameters. When the identity verification is successful and the call confidence meets the standard, the system retrieves the resources and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resources and / or execution tools to obtain the execution result.

[0149] Receive the execution result, verify the identity of the execution result and determine the confidence level of the result. When it is determined that the identity verification is successful and the execution credibility meets the standard, the task is delivered.

[0150] Or include:

[0151] The system receives the description and resource content of the execution tool to be registered sent by the intelligent agent, performs hash calculation and signature processing on the description and resource content and sends it to the blockchain so that the blockchain can sign and verify the legality of the description and resource content, compare it with the key fields of the content to obtain the registration credibility, and make a judgment on the storage of the description and resource content based on the registration credibility.

[0152] The system receives the call parameters sent by the intelligent agent, performs identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result; the call parameters are the call parameters of the resource content and / or execution tools required to execute the target task as determined by the intelligent agent.

[0153] The execution result is sent to the agent so that the agent can verify the identity of the execution result and make a confidence judgment on the result. When the identity verification is successful and the execution credibility meets the standard, the agent performs the task delivery operation on the execution result.

[0154] Furthermore, the logical instructions in the aforementioned memory 63 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0155] On the other hand, the present invention also provides a computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute a trusted model context interaction method, the method comprising:

[0156] Send the description of the execution tool and the resource content to be registered to the server. The server will then perform hash calculation and signature processing on the description of the execution tool and the resource content before sending it to the blockchain. The blockchain will then sign and verify the legality of the description of the execution tool and the resource content, and compare it using key content fields to obtain the registration credibility. Based on the registration credibility, the description of the execution tool and the resource content will be entered into the database.

[0157] The system determines the resources and / or execution tools required to perform the target task, sends call parameters representing the resources and / or execution tools to the server, so that the server can perform identity verification and call confidence judgment on the call parameters. When the identity verification is successful and the call confidence meets the standard, the system retrieves the resources and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resources and / or execution tools to obtain the execution result.

[0158] Receive the execution result, verify the identity of the execution result and determine the confidence level of the result. When it is determined that the identity verification is successful and the execution credibility meets the standard, the task is delivered.

[0159] Or include:

[0160] The system receives the description and resource content of the execution tool to be registered sent by the intelligent agent, performs hash calculation and signature processing on the description and resource content and sends it to the blockchain so that the blockchain can sign and verify the legality of the description and resource content, compare it with the key fields of the content to obtain the registration credibility, and make a judgment on the storage of the description and resource content based on the registration credibility.

[0161] The system receives the call parameters sent by the intelligent agent, performs identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result; the call parameters are the call parameters of the resource content and / or execution tools required to execute the target task as determined by the intelligent agent.

[0162] The execution result is sent to the agent so that the agent can verify the identity of the execution result and make a confidence judgment on the result. When the identity verification is successful and the execution credibility meets the standard, the agent performs the task delivery operation on the execution result.

[0163] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform a trusted model context interaction method, the method comprising:

[0164] Send the description of the execution tool and the resource content to be registered to the server. The server will then perform hash calculation and signature processing on the description of the execution tool and the resource content before sending it to the blockchain. The blockchain will then sign and verify the legality of the description of the execution tool and the resource content, and compare it using key content fields to obtain the registration credibility. Based on the registration credibility, the description of the execution tool and the resource content will be entered into the database.

[0165] The system determines the resources and / or execution tools required to perform the target task, sends call parameters representing the resources and / or execution tools to the server, so that the server can perform identity verification and call confidence judgment on the call parameters. When the identity verification is successful and the call confidence meets the standard, the system retrieves the resources and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resources and / or execution tools to obtain the execution result.

[0166] Receive the execution result, verify the identity of the execution result and determine the confidence level of the result. When it is determined that the identity verification is successful and the execution credibility meets the standard, the task is delivered.

[0167] Or include:

[0168] The system receives the description and resource content of the execution tool to be registered sent by the intelligent agent, performs hash calculation and signature processing on the description and resource content and sends it to the blockchain so that the blockchain can sign and verify the legality of the description and resource content, compare it with the key fields of the content to obtain the registration credibility, and make a judgment on the storage of the description and resource content based on the registration credibility.

[0169] The system receives the call parameters sent by the intelligent agent, performs identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result; the call parameters are the call parameters of the resource content and / or execution tools required to execute the target task as determined by the intelligent agent.

[0170] The execution result is sent to the agent so that the agent can verify the identity of the execution result and make a confidence judgment on the result. When the identity verification is successful and the execution credibility meets the standard, the agent performs the task delivery operation on the execution result.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A reliable model context interaction method, characterized in that, include: The intelligent agent sends the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration credibility, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration credibility. The intelligent agent determines the resources and / or execution tools required to execute the target task, and sends the calling parameters representing the resources and / or execution tools to the server so that the server can perform identity verification and call confidence judgment on the calling parameters. When it is determined that the identity verification is successful and the call confidence meets the standard, the intelligent agent obtains the resources and / or execution tools from the blockchain according to the calling parameters, and executes the target task according to the obtained resources and / or execution tools to obtain the execution result. The intelligent agent receives the execution result, performs identity verification and result confidence judgment on the execution result, and performs task delivery operation on the execution result when it is determined that the identity verification is successful and the execution credibility meets the standard. The step of verifying the identity and determining the confidence level of the execution result includes: The execution result is signed and its identity is verified. The execution result is signed by the server and then fed back to the intelligent agent. The data structure of the execution result is verified item by item against the data structure of the data output mode registered by the agent, and an execution credibility characterization of the execution result is generated. The method further includes: Once the signature and identity validity verification is passed and the execution credibility is within the sampling inspection score range, the task delivery operation is performed on the execution result, and the sampling inspection rate of the execution result is increased. When it is determined that the signature and identity validity verification is successful and the execution credibility is within the range of the rejection delivery score, the execution result is rejected for delivery.

2. The reliable model context interaction method according to claim 1, characterized in that, The method further includes: Before the intelligent agent sends the calling parameters representing the resource content and / or execution tool to the server, the calling parameters are signed.

3. The reliable model context interaction method according to claim 1, characterized in that, The method further includes: Receive the registration credibility of the intelligent agent's registered resource content and / or execution tools; The overall credibility is determined based on the registration credibility, invocation credibility, and execution credibility. Based on the overall credibility, the servers corresponding to the overall credibility are classified into different levels.

4. A reliable model context interaction method, characterized in that, include: The server receives the description content of the execution tool and resource content to be registered from the intelligent agent. After performing hash calculation and signature processing on the description content of the execution tool and resource content, it sends it to the blockchain so that the blockchain can sign and verify the legality of the description content of the execution tool and resource content, compare it with the key fields of the content, obtain the registration credibility, and make a judgment on the storage of the description content of the execution tool and resource content based on the registration credibility. The server receives the calling parameters sent by the intelligent agent, performs identity verification and calling confidence judgment on the calling parameters, and when it is determined that the identity verification is successful and the calling confidence meets the standard, it obtains resource content and / or execution tools from the blockchain according to the calling parameters, and executes the target task according to the obtained resource content and / or execution tools to obtain the execution result; the calling parameters are the calling parameters of resource content and / or execution tools required for executing the target task as determined by the intelligent agent; The server sends the execution result to the agent, so that the agent can perform identity verification and result confidence judgment on the execution result. When it is determined that the identity verification is successful and the execution credibility meets the standard, the agent performs the task delivery operation on the execution result. The step of verifying the identity and determining the call confidence of the call parameters includes: The call parameters are verified by signature, identity validity, and timeliness. The data structure of the calling parameters and the data structure of the agent registration are compared by field comparison to generate the calling credibility of the calling parameters.

5. A reliable model context interaction device based on the reliable model context interaction method according to any one of claims 1-3, characterized in that, include: The first registration trust module is used to send the description content of the execution tool and the resource content to be registered to the server, so that the server performs hash calculation and signature processing on the description content of the execution tool and the resource content and sends it to the blockchain, so that the blockchain performs signature and legality verification on the description content of the execution tool and the resource content, compares it with the key fields of the content, obtains the registration trust, and makes a judgment on the storage of the description content of the execution tool and the resource content based on the registration trust. The first call trust module is used to determine the resource content and / or execution tools required to execute the target task, and send call parameters representing the resource content and / or execution tools to the server so that the server can perform identity verification and call trust judgment on the call parameters. When it is determined that the identity verification is successful and the call trust is up to standard, the server retrieves the resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the retrieved resource content and / or execution tools to obtain the execution result. The first execution trust module is used to receive the execution result, perform identity verification and result confidence judgment on the execution result, and perform task delivery operation on the execution result when it is determined that the identity verification is successful and the execution trust is up to standard.

6. A trusted model context interaction device based on the trusted model context interaction method of claim 4, characterized in that, include: The second registration trust module is used to receive the description content of the execution tool and the resource content to be registered sent by the intelligent agent, perform hash calculation and signature processing on the description content of the execution tool and the resource content and send them to the blockchain, so that the blockchain can sign and verify the legality of the description content of the execution tool and the resource content, compare them using the key fields of the content, obtain the registration trust, and make a judgment on the storage of the description content of the execution tool and the resource content based on the registration trust. The second trusted module is used to receive the call parameters sent by the intelligent agent, perform identity verification and call confidence judgment on the call parameters, and when it is determined that the identity verification is successful and the call confidence meets the standard, it obtains resource content and / or execution tools from the blockchain according to the call parameters, and executes the target task according to the obtained resource content and / or execution tools to obtain the execution result. The second execution trust module is used to send the execution result to the intelligent agent so that the intelligent agent can perform identity verification and result confidence judgment on the execution result. When it is determined that the identity verification is successful and the execution trust is up to standard, the intelligent agent performs task delivery operation on the execution result.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the trusted model context interaction method as described in any one of claims 1-3, or the trusted model context interaction method as described in claim 4.

Images