Counterfeit content identifying and tracing method based on multi-agent collaboration

Abstract

The invention relates to a method and a system for identifying and tracing counterfeited content based on multi-agent collaboration. The method comprises the following steps of: acquiring to-be-identified content input by a user in a natural language dialogue or file uploading mode; the type of a task input by a user is judged through an intention recognition agent, and a corresponding detection and traceability agent is called through a modal self-adaptive routing mechanism; calling an identification agent of a corresponding mode to judge whether the to-be-identified content is generated in a counterfeit manner or not; when the content is judged to be counterfeited, analyzing a generation source of the counterfeited content through a traceability agent, and identifying a generation model category; and generating a natural language interpretation text for the detection and traceability result by an interpretation agent in combination with an expert cue word template and a counterfeit feature knowledge base. Compared with the prior art, integration and intellectualization of detection and traceability are realized. The problems of task splitting, unexplainable result, lack of traceability, complex interaction and the like of an existing detection system are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence content security and information tracing technology, and in particular to a method for identifying and tracing counterfeit content based on multi-agent collaboration. Background Technology

[0002] In recent years, with the rapid development of generative artificial intelligence (AI), AI-generated content (AIGC) has been widely used in text writing, image generation, video synthesis, and virtual human technology. However, the rapid popularization of AIGC has also brought about many security and trust issues: the spread of false information, deepfake videos, and AI-generated news and images are becoming increasingly prominent, seriously affecting the authenticity of the information ecosystem, the security of public opinion, and the reliability of social governance. Currently, although research on the detection of AI-generated content has made some progress, the following major shortcomings still exist: 1. A significant problem of modal fragmentation. Most existing detection systems are designed for a single modality, such as text generation detection, image forgery detection, or video deepfake detection. They lack a unified detection framework and interface, making it impossible to perform holistic judgment and correlation analysis of multimodal content. 2. Uninterpretable results. Traditional detection models only output "true / false" or probability values, failing to provide explanatory information such as the basis for generation, the location of forgery, or the source of evidence. This makes it difficult for users to understand the logical basis of the detection conclusions, affecting the system's credibility. 3. Weak traceability capabilities. Existing systems mostly remain at the level of "detecting whether it was generated by AI," lacking the technical capability to trace which type of model, algorithm, or data source generated the content, making it difficult to achieve true AI content responsibility tracking.

[0003] Meanwhile, technological advancements in artificial intelligence have provided new solutions for combating fake content. Multi-agent architectures possess capabilities such as task decomposition, collaborative decision-making, and result fusion, enabling collaborative information processing and interpretation in complex multimodal environments. However, existing detection systems have not yet established a unified framework based on agent collaboration, and the lack of effective connections between different detection modules leads to fragmented detection results, weak interpretability, and insufficient scalability.

[0004] Based on the above issues, how to construct an AI-generated content identification and tracing method with multimodal adaptability, interpretability, and traceability, and how to achieve task decomposition, result fusion, and continuous optimization through intelligent agent collaboration, has become a key technical challenge in the field of AI security and trusted content governance. Summary of the Invention

[0005] The purpose of this invention is to address the problems of existing detection methods relying heavily on single-modal models and lacking a unified interface and interpretation mechanism, resulting in fragmented detection results, low credibility, and difficulty in understanding. This invention provides a method for identifying and tracing forged content based on multi-agent collaboration.

[0006] The objective of this invention can be achieved through the following technical solutions: As a first aspect of the present invention, a method for identifying and tracing forged content based on multi-agent collaboration is provided, comprising the following steps: Obtain the content to be authenticated by the user through natural language dialogue or file upload; The intent recognition agent determines the type of user input task and calls the corresponding detection and tracing agent through the modality adaptive routing mechanism; Based on the modality of the input content recognized by the intent recognition agent, the corresponding modality of the discrimination agent is invoked to determine whether the content to be identified is forged or generated; When content is determined to be forged, the source of the forged content is analyzed by the source tracing agent, the generation model category is identified and the generation model confidence score is given; The interpreting agent combines expert prompt word templates and a forgery feature knowledge base to generate natural language explanation text for the detection and tracing results.

[0007] As a preferred technical solution, the intent recognition agent is based on a semantic parsing model to classify and recognize the intent of the user-input content to be identified, determine the modality and give a modality confidence vector; select the most matching modality type based on the modality confidence and pass it to the modality identification agent of the corresponding modality.

[0008] As a preferred technical solution, the identification agent includes: a text identification sub-agent, an image identification sub-agent, a video identification sub-agent, and a face identification sub-agent; After receiving confirmation of the input modality from the intent recognition agent, the corresponding identification sub-agent is invoked; each sub-agent outputs the identification result and identification confidence level of whether the content to be detected is forged based on the modality features through the feature extraction and classification model of the corresponding modality.

[0009] As a preferred technical solution, the source tracing agent infers the generation model category, version, and possible generation framework of the content based on model fingerprint recognition, statistical feature comparison, and watermark detection technology; the output includes a source tracing information set including model category, model version, model source confidence, and model inference evidence.

[0010] As a preferred technical solution, the interpreting agent receives the identification results from the multimodal identification agent and the source information set from the source tracing agent, and generates interpretable natural language analysis text based on the expert prompt word template library and the forgery feature knowledge base; the output content includes the detection reason, the basis for the anomaly, the feature analysis and the source tracing explanation.

[0011] As a preferred technical solution, the method also employs a visual intelligent agent to graphically display the detection evidence, forged areas, or text verification results, and outputs a structured and interpretable detection report; For video and image content, the visualization agent calls segmentation and object detection tools to label fake regions or abnormal frames; For text or news content, fact-checking and similar corpus comparison tools are invoked to display supporting evidence and verification links.

[0012] As a preferred technical solution, the method also employs a collaborative and learning intelligent agent. During the process of identifying and tracing counterfeit content, the system continuously monitors the operation status of modules such as intent recognition, modal adaptive routing, multimodal identification, source analysis, interpretation generation and visualization, and records operational metadata such as input modality distribution, detection link, model response time, result consistency and abnormal situations. After the task is completed, collect user feedback on the detection conclusions and traceability results, including confirmation, correction or supplementary explanation of the judgment results, and convert the feedback information into labeled data samples and store them in the feedback sample library. The collaborative and learning agent dynamically organizes the feedback sample library and generates an updated set through confidence screening, quality assessment and noise removal, which drives the continuous expansion of the forgery feature knowledge base, model fingerprint library and expert prompt word template library. After the forgery feature knowledge base is updated, the collaborative and learning agent triggers the incremental learning mechanism based on the accumulated sample size and performance evaluation index. This fine-tunes the classification model of each modality identification sub-agent, the source tracing model of the source tracing agent, and the prompt word structure of the explanation agent, so that the model parameters can absorb the feature distribution of the new forgery pattern. The collaborative and learning agents periodically evaluate the detection accuracy, source tracing stability, interpretation consistency, and visualization effectiveness of each sub-model, and optimize the model weights, modal routing thresholds, and information interaction structures between different agents based on the evaluation results.

[0013] As a second aspect of the present invention, a system for identifying and tracing counterfeit content based on multi-agent collaboration is provided. The system executes the method for identifying and tracing counterfeit content based on multi-agent collaboration as described above, including: The intent recognition agent determines the type of user input task and calls the corresponding detection and tracing agent through a modal adaptive routing mechanism; The multimodal discriminative agent, based on the input modality recognized by the intent recognition agent, calls the corresponding discriminative sub-agent to determine whether the content to be discriminated is forged; The source tracing agent analyzes the source of content that is determined to be forged, identifies the generation model category, and provides the confidence level of the generation model. The interpreting agent generates natural language explanation text for detection and tracing results based on expert prompt word templates and a knowledge base of forgery features. The visual intelligent agent graphically displays the detection evidence, forged areas, or text verification results, and outputs a structured and interpretable detection report; Collaborative and learning agents monitor task flows and coordinate the invocation order of agents for intent recognition, identification, tracing, interpretation, and visualization; user feedback data is collected and updated in the forgery feature knowledge base, and the process is optimized with prompt words.

[0014] As a preferred technical solution, the identification agent includes: a text identification sub-agent, an image identification sub-agent, a video identification sub-agent, and a face identification sub-agent; After receiving confirmation of the input modality from the intent recognition agent, the corresponding identification sub-agent is invoked; each sub-agent outputs the identification result and identification confidence level of whether the content to be detected is forged based on the modality features through the feature extraction and classification model of the corresponding modality.

[0015] As a preferred technical solution, the source tracing agent infers the generation model category, version, and possible generation framework of the content based on model fingerprint recognition, statistical feature comparison, and watermark detection technology; the output includes a source tracing information set including model category, model version, model source confidence, and model inference evidence.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1) This invention proposes a method for identifying and tracing forged content. Through the collaborative operation of an intent-recognition agent, an identification agent, a tracing agent, an interpretation agent, and a visualization agent, a closed-loop processing architecture is constructed, encompassing unified detection, source tracing, interpretation generation, and evidence presentation. This technical solution enables unified analysis and source tracing of different types of input content, significantly improving the comprehensiveness, accuracy, and automation level of the detection task, achieving integration and intelligence of detection and tracing. It effectively solves the problems of task fragmentation, uninterpretable results, lack of tracing, and complex interactions in existing detection systems.

[0017] 2) This invention possesses high interpretability and excellent user-friendliness. The interpreting agent, combining expert prompt templates and a knowledge base of forgery characteristics, transforms complex detection and tracing results into logically clear and semantically understandable natural language explanations; the visualization agent automatically generates annotations and evidence displays, achieving a graphical and traceable presentation of the detection basis. This mechanism significantly improves the transparency and credibility of the results, enhances users' understanding and trust in the detection conclusions, meets the regulatory and application needs for transparency, traceability, and verifiability in the field of forgery content governance, and provides stable and scalable technical support for content security and trustworthy review. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for identifying and tracing counterfeit content based on multi-agent collaboration according to the present invention.

[0019] Figure 2 This is a schematic diagram of the forgery content identification and tracing system based on multi-agent collaboration of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0021] Example 1 This invention proposes a multi-agent collaborative method for identifying and tracing the source of forged content, used for authenticity detection, source tracing, and interpretability analysis of multimodal content such as text, images, videos, and faces. It introduces a multi-agent collaborative mechanism, integrating intent recognition, modal adaptive routing, multimodal detection and source tracing, interpretation generation based on a large language model, and visualization, achieving unified detection, traceable analysis, and interpretable output of AI-generated content, significantly improving detection efficiency, transparency, and system scalability. The specific steps are as follows: Step 1: Obtain the content to be authenticated input by the user through natural language dialogue or file upload.

[0022] Step 2: Automatically determine the type of user input task (such as text, image, video or face detection) through the intent recognition agent, and call the corresponding detection and tracing agent through the modality adaptive routing mechanism.

[0023] Step 3: The multimodal identification agent calls the corresponding detection model based on the input modal features to determine whether the content is AI-generated. When the content is determined to be AI-generated, the source tracing agent analyzes its generation source, identifies the generation model category, version and confidence level, and realizes the traceability of the content.

[0024] Step 4: The interpreter combines expert prompt word templates and a forgery feature knowledge base to generate logically clear and highly readable natural language explanation text for the detection and tracing results.

[0025] Step 5: The visualization agent graphically displays the detection evidence, forged areas, or text verification results, and outputs a structured and interpretable detection report.

[0026] The intent recognition agent works by having the user input content to be identified via natural language dialogue or file upload. Based on a semantic parsing model, the agent performs intent classification and modality recognition on the input content, determining whether it belongs to text, image, video, or facial modality. The agent outputs a modality confidence vector. Automatically select the most matching modality type The agent then transmits the results to the corresponding detection process. It supports a context memory mechanism, allowing for dynamic updates to the recognition results during continuous dialogue.

[0027] Multimodal discriminative agent: The multimodal discriminative agent is responsible for judging whether the input content is generated by AI.

[0028] Once the intent recognition agent confirms the input modality, the multimodal identification agent calls the corresponding detection sub-agent, including: text identification sub-agent, image identification sub-agent, video identification sub-agent, or face identification sub-agent.

[0029] Each sub-agent possesses an independent feature extraction and classification model. Based on modal characteristics Output identification results And confidence level. This process enables automatic recognition and standardized output of AI-generated content under different modalities.

[0030] Source tracing agent: If the identification result indicates that the input is AI-generated forged content (i.e. The source tracing agent analyzes the content's origin. Based on model fingerprinting, statistical feature comparison, and watermark detection technologies, this agent infers the content's generation model category, version, and possible generation framework. Its output is a structured source tracing information set. This enables the traceability of AI-generated content.

[0031] Explanatory Agent: The explanatory agent is built based on a large language model and receives structured results from multimodal identification and tracing agents. That is, whether the content is forged, its confidence level, the model from which it was generated, and the evidence. Combining an expert prompt word template library and a forgery feature knowledge base, interpretable natural language analysis text is generated: The output includes the reason for detection, evidence of anomalies, feature analysis, and source tracing explanation. The interpreting agent can automatically select the interpretation style according to different modalities, supporting multiple generation modes such as professional review report style and simplified explanation style.

[0032] Visual agent: The visual agent is responsible for visually displaying the detection and interpretation results.

[0033] For video and image content, the visualization agent invokes segmentation and object detection tools to label fake regions or abnormal frames.

[0034] For text or news content, fact-checking and similar corpus comparison tools are invoked to display supporting evidence and verification links.

[0035] Ultimately, an interactive visual report is generated. It allows users to view the testing basis and key evidence.

[0036] Collaboration and Continuous Learning: Task allocation and information sharing among agents are achieved through a multi-agent coordination mechanism.

[0037] The collaborative and learning agent continuously monitors the operational status of modules such as intent recognition, modality routing, multimodal identification, source tracing analysis, interpretation generation, and visualization throughout the entire processing flow, and records operational metadata such as input modality distribution, detection chain, model response time, result consistency, and anomalies. After task execution, the system collects user feedback on detection conclusions and source tracing results, including confirmation, correction, or supplementary explanations of the judgment results, and converts this feedback into labeled data samples, storing them in a feedback sample library. The collaborative and learning agent dynamically organizes this sample library, generating high-quality update sets through confidence filtering, quality assessment, and noise removal. It is used to drive the continuous expansion of the forgery feature knowledge base, model fingerprint base, and expert prompt word template base.

[0038] After the forgery feature knowledge base is updated, the collaborative and learning agent triggers an incremental learning mechanism based on the accumulated sample size and performance evaluation metrics to improve the classification model of the identification sub-agent. Origin tracing model of origin tracing agent Furthermore, the system employs lightweight fine-tuning of the prompt word structure for interpreting agents, enabling model parameters to absorb the feature distribution of new forgery patterns and improving adaptability to novel generative models and attack methods. The training process utilizes a controllable incremental learning strategy to avoid catastrophic forgetting of historical capabilities and ensure consistency in detection output across major modalities before and after updates. Simultaneously, the collaborative and learning agents periodically evaluate the detection accuracy, source tracing stability, interpretability consistency, and visualization effectiveness of each sub-model. Based on the evaluation results, the system optimizes model weights, modality routing thresholds, and the information interaction structure between different agents, enabling the system to develop a stable and efficient agent collaboration mode during continuous operation.

[0039] Through the aforementioned closed-loop mechanism of monitoring-feedback-update-verification, the collaborative and learning agents enable the system to adaptively evolve its parameters, knowledge, and routing strategies over long-term operation. This allows the multi-agent forged content identification and tracing method proposed in this invention to continuously improve its detection performance, interpretation quality, and tracing capabilities in the face of constantly changing generation models and forgery techniques, thus forming an intelligent identification and tracing system with self-evolving characteristics. This mechanism ensures that the system can adapt to changes in new AI generation models and forgery methods, maintaining long-term detection accuracy and usability.

[0040] This invention achieves a complete processing flow from user input to content detection, source tracing, result interpretation, and evidence visualization through the collaborative work of multiple functional intelligent agents. The system consists of an intent recognition agent, a multimodal identification agent, a source tracing agent, an interpretation agent, and a visualization agent. These agents work collaboratively through task division and information exchange, thereby achieving efficient detection and reliable source tracing of multimodal forged content.

[0041] This invention has a wide range of applications, applicable to various fields and systems involving AI-generated content detection, tracing, and trust verification, with significant application value, particularly in media security, social platform governance, judicial evidence collection, brand protection, education and research, and government regulation. The method, electronic device, readable medium, and program products of this invention can all be efficiently deployed and applied in the following typical scenarios: For example, in news media or information platforms, this invention can serve as the intelligent core for backend content review. When a user uploads news articles, images, or short videos, the conversational, interpretable, unified identification and tracing agent can automatically identify the input modality, invoke the corresponding multimodal identification agent, quickly determine whether the content was generated by AI, and the tracing agent analyzes possible generation models and source channels. The system then uses the interpreting agent to generate a readability report, pointing out the detection basis and tracing evidence, and uses a visualization agent to display key forged areas or abnormal frames, thereby helping editors and reviewers identify false content and potentially misleading information in a short time, significantly improving the authenticity and credibility of news releases.

[0042] In video and social media platforms, this invention can be embedded in content review processes to detect deepfake videos or AI-generated fake faces. The system can perform localized forgery detection and temporal consistency analysis on video frame sequences, and generate user-understandable "forgery explanation" reports through an interpretive agent, such as "The video detected lip-syncing and speech asynchrony at 2.8 seconds" or "The facial texture features are suspected to be AI-generated." This mechanism significantly improves the platform's security governance capabilities and transparency.

[0043] In the fields of judicial evidence collection and cybersecurity, this invention can be used to verify the authenticity of digital evidence. Through a unified intelligent agent for identification and tracing, courts or law enforcement agencies can determine the authenticity and trace the source of image, video, audio, or document evidence, and generate structured reports and visual evidence annotations, providing technical support and credible evidence for case adjudication.

[0044] In educational and research settings, this invention can be used for AI-generated text detection, academic achievement authenticity verification, and AI-assisted source tracing of papers. The system can automatically identify AI-generated content in papers, reports, or assignments, and provide specific reasons for detection and correction suggestions, thereby maintaining academic integrity and research transparency.

[0045] Furthermore, the conversational interface of this invention also supports users initiating source tracing or verification inquiries for specific content. For example, users can directly ask, "Was this video generated by AI?" or "Which diffusion model might this image come from?" The system will automatically complete detection, source tracing, and interpretation through an intelligent agent collaboration mechanism, and return the results in the form of a natural language report. This feature enables this invention to be widely adaptable to the application needs of different fields and levels, providing an efficient, intelligent, and explainable solution for AI content governance and trustworthy review.

[0046] Example 2 As another embodiment of the present invention, this embodiment also provides a multi-agent collaborative identification and tracing system, such as... Figure 2 As shown, the system includes: Intent Recognition Agent: Users input content to be identified via natural language dialogue or file upload. The intent recognition agent, based on a semantic parsing model, performs intent classification and modality recognition on the input content, determining whether it belongs to text, image, video, or facial modality. The agent outputs a modality confidence vector. Automatically select the most matching modality type The agent then transmits the results to the corresponding detection process. It supports a context memory mechanism, allowing for dynamic updates to the recognition results during continuous dialogue.

[0047] The system employs multimodal identification and source tracing agents. The identification agents include text identification sub-agents, image identification sub-agents, video identification sub-agents, and face identification sub-agents. Based on the input modality identified by the intent recognition agent, the corresponding identification sub-agent is invoked to determine whether the content to be identified is forged. For content determined to be forged, the source of its generation is analyzed, the generation model category is identified, and the generation model confidence score is provided.

[0048] Source tracing agent: If the identification result indicates that the input is AI-generated forged content, the source tracing agent will analyze its generation source. Based on model fingerprint recognition, statistical feature comparison, and watermark detection technology, this agent infers the content generation model category, version, and possible generation framework, and outputs a structured source tracing information set.

[0049] The explanation agent receives structured results from a multimodal identification and source tracing agent. Based on expert prompt word templates and a forgery feature knowledge base, it generates natural language explanation text for the detection and source tracing results. The output includes the reason for detection, evidence of anomalies, feature analysis, and source tracing explanation. The explanation agent can automatically select an explanation style according to different modalities, supporting multiple generation modes such as professional review report style and simplified explanation style.

[0050] The visual intelligent agent graphically displays the detection evidence, forged areas, or text verification results, and outputs a structured and interpretable detection report.

[0051] The collaborative and learning intelligent agent monitors the task flow, coordinates the calling order of the intent recognition, identification, tracing, interpretation and visualization intelligent agents; and collects user feedback data into the forgery feature knowledge base to update and optimize the prompt word process.

[0052] Visual intelligent agents include document, image, and video annotation tools. If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, 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 this 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.

[0053] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for identifying and tracing forged content based on multi-agent collaboration, characterized by the following steps: include: Obtain the content to be authenticated by the user through natural language dialogue or file upload; The intent recognition agent determines the type of user input task and calls the corresponding detection and tracing agent through the modality adaptive routing mechanism; Based on the modality of the input content recognized by the intent recognition agent, the corresponding modality of the discrimination agent is invoked to determine whether the content to be identified is forged or generated; When content is determined to be forged, the source of the forged content is analyzed by the source tracing agent, the generation model category is identified and the generation model confidence score is given; The interpreting agent combines expert prompt word templates and a forgery feature knowledge base to generate natural language explanation text for the detection and tracing results.

2. The method for identifying and tracing forged content based on multi-agent collaboration according to claim 1, characterized in that, The intent recognition agent is based on a semantic parsing model. It performs intent classification and modality recognition on the user input content to be identified, determines the modality, and provides a modality confidence vector. Based on the modality confidence, it selects the most matching modality type and passes it to the modality identification agent of the corresponding modality.

3. The method for identifying and tracing forged content based on multi-agent collaboration according to claim 1, characterized in that, The identification agent includes: a text identification sub-agent, an image identification sub-agent, a video identification sub-agent, and a face identification sub-agent; After receiving confirmation of the input modality from the intent recognition agent, the corresponding identification sub-agent is invoked; each sub-agent outputs the identification result and identification confidence level of whether the content to be detected is forged based on the modality features through the feature extraction and classification model of the corresponding modality.

4. The method for identifying and tracing forged content based on multi-agent collaboration according to claim 1, characterized in that, The source tracing agent, based on model fingerprint recognition, statistical feature comparison, and watermark detection technologies, infers the content generation model category, version, and possible generation framework; the output includes a source tracing information set including model category, model version, model source confidence, and model inference evidence.

5. The method for identifying and tracing forged content based on multi-agent collaboration according to claim 1, characterized in that, The interpreting agent receives the identification results from the multimodal identification agent and the source information set from the source tracing agent. Based on the expert prompt word template library and the forgery feature knowledge base, it generates interpretable natural language analysis text. The output includes the detection reason, the basis for the anomaly, feature analysis, and source tracing explanation.

6. The method for identifying and tracing forged content based on multi-agent collaboration according to claim 1, characterized in that, The method also employs a visual intelligent agent to graphically display the detected evidence, forged regions, or text verification results, and outputs a structured and interpretable detection report; For video and image content, the visualization agent calls segmentation and object detection tools to label fake regions or abnormal frames; For text or news content, fact-checking and similar corpus comparison tools are invoked to display supporting evidence and verification links.

7. The method for identifying and tracing counterfeit content based on multi-agent collaboration according to claim 6, characterized in that, The method also employs collaborative and learning intelligent agents. During the process of identifying and tracing counterfeit content, the system continuously monitors the operation status of modules such as intent recognition, modal adaptive routing, multimodal identification, source analysis, interpretation generation and visualization, and records operational metadata such as input modality distribution, detection link, model response time, result consistency and abnormal situations. After the task is completed, collect user feedback on the detection conclusions and traceability results, including confirmation, correction or supplementary explanation of the judgment results, and convert the feedback information into labeled data samples and store them in the feedback sample library. The collaborative and learning agent dynamically organizes the feedback sample library and generates an updated set through confidence screening, quality assessment and noise removal, which drives the continuous expansion of the forgery feature knowledge base, model fingerprint library and expert prompt word template library. After the forgery feature knowledge base is updated, the collaborative and learning agent triggers the incremental learning mechanism based on the accumulated sample size and performance evaluation index. This fine-tunes the classification model of each modality identification sub-agent, the source tracing model of the source tracing agent, and the prompt word structure of the explanation agent, so that the model parameters can absorb the feature distribution of the new forgery pattern. The collaborative and learning agents periodically evaluate the detection accuracy, source tracing stability, interpretation consistency, and visualization effectiveness of each sub-model, and optimize the model weights, modal routing thresholds, and information interaction structures between different agents based on the evaluation results.

8. A system for identifying and tracing counterfeit content based on multi-agent collaboration, characterized in that, The system executes the method for identifying and tracing forged content based on multi-agent collaboration as described in any one of claims 1-7, including: The intent recognition agent determines the type of user input task and calls the corresponding detection and tracing agent through a modal adaptive routing mechanism; The multimodal discriminative agent, based on the input modality recognized by the intent recognition agent, calls the corresponding discriminative sub-agent to determine whether the content to be discriminated is forged; The source tracing agent analyzes the source of content that is determined to be forged, identifies the generation model category, and provides the confidence level of the generation model. The interpreting agent generates natural language explanation text for detection and tracing results based on expert prompt word templates and a knowledge base of forgery features. The visual intelligent agent graphically displays the detection evidence, forged areas, or text verification results, and outputs a structured and interpretable detection report; Collaborative and learning agents monitor task flows and coordinate the invocation order of agents for intent recognition, identification, tracing, interpretation, and visualization; user feedback data is collected and updated in the forgery feature knowledge base, and the process is optimized with prompt words.

9. A system for identifying and tracing counterfeit content based on multi-agent collaboration as described in claim 8, characterized in that, The identification agent includes: a text identification sub-agent, an image identification sub-agent, a video identification sub-agent, and a face identification sub-agent; After receiving confirmation of the input modality from the intent recognition agent, the corresponding identification sub-agent is invoked; each sub-agent outputs the identification result and identification confidence level of whether the content to be detected is forged based on the modality features through the feature extraction and classification model of the corresponding modality.

10. A system for identifying and tracing counterfeit content based on multi-agent collaboration as described in claim 8, characterized in that, The source tracing agent, based on model fingerprint recognition, statistical feature comparison, and watermark detection technologies, infers the content generation model category, version, and possible generation framework; the output includes a source tracing information set including model category, model version, model source confidence, and model inference evidence.

Images