| Project Number | Project Type | Project Title | Scope | Project Status |
| P2961 | Entity | Guide for an Architectural Framework and Application for Collaborative Edge Computing | This guide defines a machine learning framework that allows a computing task to be decomposed and distributed across edge and cloud nodes. This guide provides a blueprint for data usage, model learning, and computing collaboration in edge computing environments while meeting latency, privacy, security, and regulatory requirements. It defines the architectural framework and application guidelines for collaborative edge computing, including 1) description and definition of collaborative edge computing, 2) the types of collaborative edge computing, 3) the application scenarios to which each type applies, and 4) performance evaluation of collaborative edge computing in the real application system. | RevCom Agenda(21 Oct 2026) |
| P3157 | Entity | Recommended Practice for Vulnerability Test for Machine Learning Models for Computer Vision Applications | This recommended practice provides a framework for vulnerability tests for machine learning models in the computer vision domain. The document covers the following areas:
– definitions of vulnerabilities for machine learning models and their training processes, |
RevCom Agenda(21 Oct 2026) |
| P3142 | Entity | Recommended Practice on Distributed Training and Inference for Large-scale Deep Learning Models | This recommended practice specifies principles, approaches, and key performance indicators for distributed training and inference of large-scale deep learning models. | Draft Development |
| P3181 | Entity | Standard for Trusted Environment Based Cryptographic Computing | This standard defines trusted environment based cryptographic computing systems including participants and computation nodes. Furthermore, the standard formulates a technical framework for trusted environment based cryptographic computing systems and guides the design, development, testing, and maintenance of such systems. The standard specifies the function and security requirements, as well as the computing procedure. | Draft Development |
| P3193 | Entity | Recommended Practice on Large-scale Pre-trained Deep Learning Model Application Framework | This recommended practice provides a framework for the use of large-scale pre-trained deep learning models, including approaches, taxonomies, related roles, and activities, software toolkits, application ability key performance indicators as well as assessment means. | SA Ballot: Recirculation |
| P3342 | Entity | Standard for Functional Requirements of Toolchain for Artificial Intelligence Model Deployment on Edge Devices | This standard outlines the toolchain for deploying artificial intelligence (AI) models on edge devices and specifies the functional requirements for this process. The standard covers key areas such as frontend adaptation, model compression, graph optimization, backend adaptation, compiling optimization, and runtime optimization. It is relevant to all stages of AI model deployment on edge devices, including design, development, promotion, and application. | Draft Development |
| P3338 | Entity | Guide for Framework for Data Contribution Evaluation in Federated Machine Learning | This guide provides data contribution measurement methods for federated machine learning. The document provides guidance with respect to data trading feasibility for federated machine learning. The guide describes three main aspects: 1) Framework for data contribution evaluation of federated machine learning, 2) specific scenario under data contribution, 3) measurement methods for data contribution of federated machine learning. | Draft Development |
| P3347 | Individual | Standard for Description Schemas for Neural Network Architectures | This standard defines a unified text approach to describing neural network architectures. This standard specifies data requirements, coding formats of neurons in neural networks, schemas for neural network architectures, and application scenarios. | SA Ballot: Comment Resolution |
| P3354 | Entity | Trial-Use Recommended Practice for 3D Object Reconstruction from Multi-view Images Based on Deep Learning | This recommended practice specifies the general requirements of three dimensional (3D) object reconstruction based on deep learning, including terminology, requirements for data format, functional processes, application scenarios and evaluating, metrics for evaluating the processes and effect of 3D object reconstruction. | Draft Development |
| P3395 | Individual | Standard for the Implementation of Safeguards, Controls, and Preventive Techniques for Artificial Intelligence (AI) Models | This standard addresses Artificial Intelligence (AI) risks such as malicious use, AI race, organizational risks, and rogue AI’s. The standard defines the relevant terminology, potential applications, and a risk abatement strategy for the behavior of AI models. The standard does not detail these foundational applications, but describes them in general. | Draft Development |
| P3394 | Individual | Standard for Large Language Model Agent Interface | This standard defines natural language interfaces that facilitate communication between Large Language Model (LLM) applications, agents, and human users. The standard defines a set of protocols and guidelines that enable applications and agents to effectively communicate with LLM enabled Agents. Thereby, the standard enables seamless interactions between multiple applications and agents. The standard covers a wide range of aspects related to LLM usage and application, including but not limited to API syntax and semantics, voice and text format, conversational flow, prompt engineering integration, LLM chain of thoughts integration, and API endpoint configuration, authentication and authorization for LLM plugins. | SA Ballot: Invitation |
| P3396 | Individual | Recommended Practice for Defining and Evaluating Artificial Intelligence (AI) Risk, Safety, Trustworthiness, and Responsibility | This recommended practice provides a comprehensive framework for understanding, defining, and evaluating AI risks, AI safety, AI trustworthiness, and AI responsibility in order to address and manage them while preserving the benefits of innovation. This recommended practice takes into consideration the global context, advocating for responsible AI adoption, governance, and collaboration. The recommended practice offers a principles-based framework that looks at the role of AI in information generation, decision-making, human agency, and responsibilities associated with AI usage while considering the full life cycle of AI application development, deployment, and operation. | Draft Development |
| P3398 | Entity | Recommended Practice for Generative Pre-trained Transformer Empowered Software Engineering Life Cycle | For the software engineering life cycle empowered by Generative Pre-trained Transformer (GPT) this recommended practice specifies: • a description and definitions, • the competency in each stage applying GPT, • levels of intelligence empowered by GPT, • intelligence level evaluation methods. A generative pre-trained transformer (GPT) is a type of neural network architecture used in natural language processing (NLP) applications. GPTs are transformational self-supervised NLP models that can generate human-like language and fine-tune to various downstream tasks, enabled by their pre-training approach and large transformer architectures. |
RevCom Agenda(22 Sep 2026) |
| P3403 | Entity | Recommended Practice on Data Processing for Training Large Language Models | This recommended practice provides a data processing framework for training large language models, refining the relevant terms and definitions. The recommended practice also provides data processing processes, methods, characteristics and performance evaluation during the pre-training and fine-tuning stages. |
Draft Development |
| P3423 | Entity | Standard for End-to-end Framework Design for Autonomous Driving | This standard addresses the evaluation of safety, explainability, and stability of algorithms implementing autonomous driving levels, as defined by SAE J3016. The standard specifies the minimum requirements for experimental setups, data flow charts, algorithm pipelines, and automated quantitative evaluation metrics for autonomous driving algorithms that use available Artificial Intelligence (AI) foundation models.
Because of the rapid pace of commercialization of large AI foundation models, this standard specifies a software escrow service for proprietary code and automated processes for algorithms that use AI foundation models in the driving scenario. |
Draft Development |
| P3417 | Entity | Standard for Requirements of Differential Privacy-based Personal Information Protection | This standard specifies the architecture and technical requirements of differential privacy for personal information protection in Artificial Intelligence model training. Differential privacy provides a framework for ensuring the privacy of individuals in datasets by allowing data to be analyzed without revealing sensitive information about any specific individual. The requirements consist of data security, algorithm security and scenario-based system security of differential privacy technology used in model training process. This standard also provides classification methodologies of differential privacy technology. | SA Ballot: Comment Resolution |
| P3419 | Individual | Standard for Large Language Model Evaluation | This standard establishes a comprehensive set of criteria for the evaluation of Large Language Models (LLMs) and extends to multimodal models. Primarily, the standard provides a multi-dimensional evaluation framework for the users, teams, and enterprises who evaluate or use LLM technology or models. The standard describes a three-dimensional evaluation framework anchored in the principles of “versatility, intelligence, efficiency, and safety.” This framework is categorized under “capability-task-metrics” and encompasses evaluations at different stages, including foundation models, pre-training algorithms, and fine-tuning algorithms. The standard provides objective and subjective viewpoints and offers an extensive understanding of LLMs’ multifaceted capabilities and potentialities. This comprehensive understanding greatly assists researchers and developers, as well as downstream users. Furthermore, the open-source evaluation platform operations embedded in the framework help stimulating innovation in model and algorithm research and drive forward their industrial applications. | Draft Development |
| P3429 | Individual | Recommended Practices for Levels of Artificial Intelligence Generated Content Technologies | This recommended practice provides an overview of Artificial Intelligence Generated Content (AIGC) technologies, levels of those AICG technologies and defines the following:
• Recommended rules for levels of AIGC technologies; |
SA Ballot: Ballot |
| P3426 | Individual | Standard for Defining and Measuring the Capabilities of AI Foundation Models | This standard defines and provides criteria to measure the capabilities of foundation models. The standard focuses on measurable and objective capabilities such as perception-oriented capabilities, cognitive capabilities, and learning capabilities. The standard provides guidelines for evaluating these capabilities, drawing inspiration from human intelligence quotient (IQ) definition and measurement, and includes:
– Definition of key capabilities pertinent to foundation models. |
Draft Development |
| P3430 | Individual | Recommended Practice for A Holistic Framework for AI Foundation Models | This recommended practice provides a holistic framework for AI foundation models. The framework allows sharing of the common computing resources, data and infrastructure in the usage(s) and application(s) of AI Foundation models. The detailed technical requirements for the holistic framework are provided. Typical use cases illustrate the use of the holistic framework for AI foundation models. | SA Ballot: Ballot |
| P3427 | Entity | Standard for Federated Machine Learning of Semantic Information Agents | This standard defines a comprehensive framework for federated machine learning of semantic information agents. It targets two primary layers:
• Information Protocol Layer: This layer focuses on binary compression and protocol-level data structures for federated information exchange, with an emphasis on efficiency and security. • Semantic Layer: This layer focuses on the meaning and interpretation of information through domain-specific languages (DSLs) that are human-understandable. The standard also provides rule-based guidance and optimization strategies for different nodes within federated learning networks, addressing role definitions for agents, reward mechanisms, and multi-agent system regulations. |
SA Ballot: Recirculation |
| P3428 | Individual | Standard for Large Language Model Agents for AI-powered Education | This standard provides:
1. Agent Components: The building blocks and architectural elements that constitute an educational Large Language Model (LLM) agent. |
Draft Development |
| P3445 | Entity | Standard for Privacy Engineering Maturity Model | This standard defines a privacy engineering maturity model, including privacy engineering capability domains, maturity levels and the corresponding requirements. The standard also includes specific considerations for privacy engineering in the entire life cycle of Artificial Intelligence products, systems and services. | RevCom Agenda(21 Oct 2026) |
| P3447 | Entity | System Architecture and General Requirements for Pre-trained Large Language Models (LLM) Applications in the Smart Home Industry | This standard provides definitions, terms, frameworks, and general requirements for systems that apply pre-trained large language models (LLM) in the smart home industry. It covers foundational platforms, model capabilities, scenario applications, and safety requirements. | Draft Development |
| P3462 | Individual | Recommended Practice for Using Safety by Design in Generative Models to Prioritize Child Safety | This recommended practice follows a safety by design approach by providing recommendations for developing, deploying, and maintaining generative artificial intelligence models with adequate safeguards against child sexual abuse. The document provides a framework that expands child safety to the entire lifecycle of machine learning (ML): development, deployment, and maintenance stages. Each part in the process includes opportunities to prioritize child safety, regardless of data modality (i.e. text, image, video, audio) or whether an organization releases its technology as closed source or open source, or some release option between these two. | RevCom Agenda(22 Sep 2026) |
| P3378.7 | Entity | Standard for Technical Requirements for Large-Scale Deep Learning Models in Power Industry | This standard provides a reference architecture for large-scale deep learning models in power industry. The reference architecture includes resource pools, architectures, tools, data resources, models, interfaces, industry applications, and service platforms. This standard specifies requirements for scenario service capabilities, general capabilities, and application maturity, which enables to evaluate the technical maturity and stability of large-scale deep learning models in power industry and determine whether large-scale deep learning models can be smoothly and effectively integrated into existing operations, processes, and systems. This standard provides a reference for third-party power industry researchers as well as regulatory and evaluation institutions to conduct capacity assessments of large-scale deep learning models in power industry. |
RevCom Agenda(21 Oct 2026) |
| P3378.6 | Entity | Standard for Large-Scale Deep Learning Models for Coding | This standard applies to enterprises in the process of research and development, evaluation, and certification of large-scale deep learning models for coding. The standard specifies a technical architecture and application requirements for large-scale deep learning models for coding and provides a reference for the construction and improvement of the capabilities and provides guidance for the selection of technology for large-scale deep learning models for coding. | SA Ballot: Pre-Ballot |
| P3481 | Entity | Standard for the Functional Requirements for Cybersecurity-Specific Large Language Models | This standard defines a cybersecurity-specific Large Language Models (LLM) architecture for assisting personnel in analyzing cybersecurity related data. The standard specifies a framework, security goals, threat handling processes, and requirements on the deployment of the model. | Draft Development |
| P3489 | Entity | Standard for Test Technical Requirements of Image Recognition Algorithms for Power Line Hidden Danger Monitoring | This standard applies to testing of Artificial Intelligence (AI)-based image recognition that is used to monitor power lines for hidden dangers. The standard provides a reference for sample set composition, test index setting, test result judgment, and test environment construction. | SA Ballot: Ballot |
| P3484 | Individual | Recommended Practice for an Artificial Intelligence Imaging and Data Process Model for Sustainable Textile Value Chain Operations | The recommended practice outlines a process model that leverages artificial intelligence imaging and data processing technologies for improved emissions information reporting. It provides textile producers with comprehensive data on fabrics, including their origins and metadata. This approach enables stakeholders to purchase only what is needed for specific product cycles, thereby increasing efficiency in supply chain operations and allowing for more accurate emissions reporting. This technology also provides traceability for end users by offering end-to-end information, facilitating increased sustainability by informing the waste typically involved in the textile to garments manufacturing processes. The practice further encourages recording inventory both used and unused to be included in the product life cycle. | Draft Development |
| P3497 | Entity | Standard for Performance Benchmarking of Artificial Intelligence for Database Management Systems for Structured Query Language-based Databases | This standard defines a methodology for performance testing of Artificial Intelligence (AI) for Database Management Systems (DMS) for Structured Query Language (SQL)-based Databases and provides a reference framework for relevant organizations and institutions.
The performance metrics encompass not only traditional database metrics like response times, resource utilization (CPU, memory, storage), and system robustness but also AI-specific metrics, such as convergence speed and model training time. |
Draft Development |
| P7999 | Individual | Standard for Integrating Organizational Ethics Oversight in Projects and Processes Involving Artificial Intelligence | This standard provides a base framework to determine the individual qualification of Artificial Intelligence (AI) ethics oversight actors and organizational certification of AI developers using existing ethically-aligned design standards developed for AI systems. It furthermore outlines best practices based on these standards for integrating ethics oversight in organizations of any scale for projects that use or develop AI and AI systems (at any stage of the AI system lifecycle). | Draft Development |
| P7999.1 | Individual | Standard for Integrating Organizational Ethics Oversight in Projects and Processes Involving Artificial Intelligence – Qualification of Individuals | This standard defines a set of metrics that determine the certification criteria that qualify individuals to perform Artificial Intelligence (AI) ethics oversight functions in organizational settings, and recommendations for the establishment of certification programs that enable individuals to specialize in specific aspects of AI ethics oversight (e.g., Large Language Models [LLMs], autonomous driving systems, generative images/videos). | Draft Development |
| P7999.2 | Individual | Standard for Integrating Organizational Ethics Oversight in Projects and Processes Involving Artificial Intelligence – Organizational Certification | This standard provides a set of metrics that can be used to certify that companies, manufacturers, and other organizational bodies employ a set of standard operating procedures that facilitate the effective deployment of an Artificial Intelligence (AI) ethics oversight function at all stages of the AI system lifecycle. These metrics include auditable elements of quality management systems utilized by companies, manufacturers, and other organizational bodies (which do not replace other quality management system conformance criteria), and post-market surveillance feedback to support ethically-aligned design control processes. | Draft Development |
| P3507 | Entity | Standard for Multimodal Large Language Models Platform for Automotive Services | This standard provides an architecture framework and functionalities related to automotive Multimodal Large Language Model (MLLM) platforms. Furthermore, the standard defines interfaces between vertical applications and automotive MLLM platforms.
The standard defines procedures for building MLLMs specifically for automotive applications. Thereby, it establishes the norms for constructing an ecosystem of service agents by defining the Application Programming Interface (API) input and output specifications. Furthermore, the standard sets forth testing and evaluation protocols and specifications. |
Draft Development |
| P3511 | Entity | Standard for Risk Management Guidelines for Generative Artificial Intelligence Systems | The standard defines guidelines on managing risk faced by organizations during the development and application of generative Artificial Intelligence (AI). Thereby, the standard assists organizations integrating risk management into generative AI systems or services lifecycle. The standard describes risk definition, risk evaluation, and risk treatment for the effective risk management implementation of generative AI. | Draft Development |
| P3514 | Individual | Recommended Practice for Levels of Artificial Intelligence Capabilities for AI Entities and Applications | This recommended practice defines levels of capabilities of Artificial Intelligence (AI). The capability levels can be used to classify AI entities after benchmarking and evaluation. Evaluation criteria and specific benchmarks are provided for common capabilities per typical application scenarios/domains (e.g., AI used in a personal computer, in a mobile phone, or by robots). | Draft Development |
| P3526 | Entity | Guide for Threat Intelligence Retrieval Framework Based on Large Language Model | This guide proposes a technical framework for using Large Language Model (LLM) technology for threat intelligence retrieval, including reference architecture, workflow, and technical implementations. LLM-based threat intelligence retrieval can be used to provide real-time information retrieval for users in cybersecurity operations and to offer threat intelligence retrieval services for cybersecurity products. It includes four parts: component layer, core layer, service layer, and maintenance management. The workflow of LLM-based threat intelligence retrieval consists of four steps: retrieval input, semantic parsing, retrieval execution, and result output. Implementations are proposed based on the functionality of each layer. | Draft Development |
| P3540 | Entity | Standard for Technical Requirements of Artificial Intelligence Personal Computer | This standard defines technical requirements for Artificial Intelligence Personal Computers (AIPCs). It includes a reference architecture, hardware, software, application interfaces, and security requirements. This standard is intended to provide guidance for the design, development, testing, deployment, and maintenance of AIPCs.
The AIPC is designed to support a broad range of AI tasks, including: The design goals of an AIPC are to provide a platform that can meet everyday computing needs while effectively supporting AI applications. This design takes into account current technological trends and looks towards future developments, aiming for a balance between performance, efficiency, and user experience. |
Draft Development |
| P3579 | Individual | Standard for Artificial Intelligence Applied to Time Series | This standard defines the fundamentals, requirements, and methodologies for applying Artificial Intelligence Systems (AIS) to time series modeling and analysis. It specifies processes for data acquisition, preprocessing, model training, fine-tuning, evaluation, and deployment of artificial intelligence models. This standard excludes any methods of time series modeling and analysis that do not rely on AIS. | Draft Development |
| P3584 | Entity | Standard for Assessment Method for Carbon Emission Measurement of Artificial Intelligence (AI) Model Training | This standard defines the scope of assessment, system boundaries, and parameter data required for evaluating the carbon emissions generated during the training of artificial intelligence (AI) models. Additionally, it elaborates on the specific assessment methodology for these carbon emissions. It applies to the carbon emission assessment of the entire AI model training process. | Draft Development |
| P3551 | Entity | Standard for Framework for Electric Power On-Device Large Language Model Evaluation | This standard is concerned with large language models (LLMs) that enhance the adaptability to hardware and power efficiency of electric power smart terminals. The standard establishes an evaluation framework for electric power on-device LLMs (OD-LLMs) that covers evaluation indicators and corresponding evaluation methods and processes. This standard defines requirements for the evaluation of OD-LLMs with respect to scenario application, model lightweighting, hardware adaptation, and user quality of experience. The standard also addresses the design and development of OD-LLMs. | Draft Development |
| P3591 | Individual | Standard for Fair Decision Making Through Causal Analysis | This standard describes how to perform causal fairness analysis to make fairer decisions in various high-stakes applications (e. g. credit, employment, education) that are more likely to be compliant with a country’s antidiscrimination laws and regulations. It provides a standardized fairness model that encodes knowledge and assumptions about how to map the causal relationships between different variables such as a protected class (e. g. race, gender) and an outcome. The document provides the reader with a standardized language for directly translating concepts among the law, causal inference, and supervised machine learning. The standard provides criteria for selecting which variables to include in a machine learning model, how to train and deploy the model to make fairer predictions and decisions, as well as how to evaluate the model to determine the likelihood for illegal discrimination. However, specific algorithms for debiasing the data in a machine learning model to help ensure fairness are not covered in this standard. Finally, this standard is designed to be focused solely on legal compliance. | SA Ballot: Invitation |
| P3704 | Entity | Standard for Evaluation Methods for the Capabilities of Embodied Intelligence | This standard provides a comprehensive framework for evaluating embodied intelligence capabilities and addresses evaluation requirements through the introduction of three hierarchical levels:
• foundational module capabilities, This standard defines functional requirements such as perception, decision-making and planning, motion control, manipulation, interaction, physical performance, and safety, and provides specific evaluation methods. The standard applies to product developers and evaluation institutions focusing on embodied intelligence, offering a reference for the design and assessment of embodied intelligence capabilities. |
Draft Development |
| P3709 | Individual | Standard for Framework and Technical Requirements of Agentic Artificial Intelligence (AI) | This standard specifies a technical framework for Agentic Artificial Intelligence (AI). The framework addresses commercial utilization and defines specific technical requirements. | Draft Development |
| P3708 | Entity | Standard for Application Specifications of Poultry Industry Intelligence Platforms | This standard defines the framework and functional requirements for the poultry industry intelligence platforms powered by artificial intelligence (AI). The framework is designed to support the integration of AI technologies with Internet of Things (IoT) infrastructure, multi-source data processing, intelligent models, and interactive services across various breeding industry domains. It includes the following layers: IoT infrastructure, middleware interaction, data management, AI model integration, application services, operations and maintenance, and security.
The standard also provides evaluation methodologies for assessing AI effectiveness in breeding industry scenarios, promoting interoperability, scalability, and sustainable development of AI-based breeding industry intelligence Platforms. |
Draft Development |
| P3707 | Entity | Standard for Embodied Artificial Intelligence Application Framework in the Elderly Care Industry | This standard defines a reference framework and core technical requirements for embodied artificial intelligence (E-AI) applications in the elderly care industry. It specifically addresses systems with physical embodiment and environmental interaction capabilities, including intelligent service robots, E-AI-driven health monitoring devices, adaptive human-robot collaboration platforms, context-aware decision support systems, and integrated operational infrastructures for elderly care institutions. The framework emphasizes the design of E-AI systems through hardware-software co-development, multi-modal data fusion, real-time environment perception, ethical safety protocols, and seamless integration with human-centric workflows. The standard covers the following key dimensions: • Embodied Functional Capabilities (e.g., environmental perception, tactile interaction, autonomous navigation, adaptive service execution) • Safety and Risk Governance (e.g., collision avoidance, emergency response, real-time anomaly detection) • Data and Privacy Protection (with emphasis on sensor data lifecycle management) • Human-Robot Interface Standards (including voice, gesture, and haptic interaction) • Ethical Considerations (e.g., transparency in decision-making, accountability for embodied actions) • Interoperability with legacy elderly care IT ecosystems The standard also provides evaluation metrics for embodied system performance (e.g., task completion accuracy, environmental adaptability), user trust, regulatory alignment, and societal impact. |
Draft Development |
| P3711 | Individual | Standard for Performance Benchmarking for Artificial Intelligence (AI) in Video Database Management Systems (DMS) | This standard defines a methodology for performance benchmarking of Artificial Intelligence (AI) for Database Management Systems (DMS) for Video Databases and provides a reference framework for relevant organizations and institutions. The performance metrics encompass traditional database metrics like response times, resource utilization (CPU, memory, storage), and system robustness, also include AI-specific metrics, such as trajectory speed and trajectory speed accuracy rate. |
Draft Development |
| P3410.1 | Entity | Standard for Constructing Large Scale Financial Risk Management Models | This standard specifies the technical requirements for the construction of Large Scale Artificial Intelligence (AI) Models in Financial Risk Management (FRM). It includes data collection and processing, model architecture design, training methodology design, evaluation metric selection. The standard also defines the application of pre-trained models to downstream financial risk assessment tasks. This standard applies to financial institutions and suppliers involved in the development and deployment of large scale financial risk models. It primarily focuses on loan default risk assessment in lending business. For people working on other risk types or other financial sectors, it provides a reference to build corresponding large scale FRM models. | SA Ballot: Invitation |
| P3723 | Entity | Standard for Intelligent Edge Device Data Processing Framework for Electric Power | This standard specifies a framework for data processing on intelligent edge devices in the field of Electric Power. It defines the architectural components and technical processes involved in the full data lifecycle at the edge, including data acquisition, preprocessing, modeling, scheduling, real-time analytics, data transmission, and security. The standard introduces the concept of data flywheel, enabling continuous feedback and self-optimization of data workflows within edge environments. | Draft Development |
| P3753 | Entity | Standard for Artificial Intelligence Reinforcement Learning System Architecture and Requirements | This standard specifies the architecture, interfaces and general requirements for Reinforcement Learning Systems. It also defines the common functions and data formats for building the reinforcement learning environment. | Draft Development |
| P3752 | Entity | Recommended Practice for the Application of Reinforcement Learning Methods in Power Dispatching and Management | This document provides recommended practices for applying reinforcement learning (RL) to power dispatch and energy management. It covers the establishment of RL models, training sample design, optimization implementation, and economic benefit evaluation, guided by the principles of safety, reliability, stability, and controllability.
It offers guidance on integrating operational constraints into model design and on using simulated and actual data during training to help ensure that deployed RL agents do not compromise system safety. Applicable to power systems with diverse controllable devices (e.g., thermal units, flexible loads), this document supports the training, testing, and benchmarking of models. The objective is to enhance operational safety and reliability across regional grids, distribution networks, and microgrids. |
Draft Development |
| P3762 | Entity | Standard for Technical Requirements for Heterogeneous Hybrid Inference Systems | This standard defines the scope of heterogeneous hybrid inference systems, encompassing key concepts, system architecture, and interoperability across hardware abstraction layers and software frameworks. It provides a comprehensive framework covering hardware/software interfaces, distributed memory architectures, cross-hardware parallel scheduling, task partitioning during inference, performance benchmarking, and the requirements for integrated toolchains that support heterogeneous hybrid inference systems. | Draft Development |
| P3796 | Entity | Standard for Binary Compliance Requirements of Industrial Vision Inspection Systems in Smart Manufacturing of Household Appliances | This standard specifies a binary compliance checklist for industrial vision inspection systems used in the smart manufacturing of household appliances. It defines clear, testable requirements for system functions and performance in a yes/no format to enable immediate applicability across stakeholders. The document covers: 1. Core functional capabilities (e.g., surface defect detection, assembly verification, dimensional checks). 2. Performance metrics with defined thresholds (e.g., minimum resolution, false reject/accept rates). 3. Interoperability requirements (e.g., Application Programming Interface (API) schema, data exchange formats). 4. Compliance testing procedures enabling pass/fail assessment. |
Draft Development |
| P3786 | Entity | Guide for Classification of Artificial intelligence Application Scenarios in the Operation and Maintenance of Electric Power Equipment | This guide applies to the classification of artificial intelligence (AI) application scenarios in the operation and maintenance of electric power equipment, with a primary focus on power generation, transmission and distribution. It provides guidance for power grid enterprises, power generation enterprises, equipment manufacturers, and relevant research institutions to identify, classify, and apply AI technologies in this field. The guide includes a classification framework for artificial intelligence in the operation and maintenance of electrical equipment, as well as specific application scenarios and the basic technical requirements for implementing AI in these contexts. |
Draft Development |
| P3777 | Entity | Standard for Benchmarking and Performance Metrics of Artificial Intelligence (AI) Agents | This standard establishes a unified framework for benchmarking Artificial Intelligence (AI) agents, including autonomous, collaborative, and task-specific agents. It defines core performance metrics, evaluation protocols, and reporting requirements to enable transparent, reproducible, and comparable assessment of AI agent capacities, capabilities, and performance. The framework supports multiple classification criteria for AI agents and emphasizes objective, binary evaluation measures wherever possible. Benchmarking dimensions may include efficiency, robustness, adaptability, ethical compliance, and interoperability options without prescribing agent-to-agent interoperation. | Draft Development |
| P3936 | Entity | Standard for Technical Requirements of Audio Large Language Models | This standard specifies the key terminology, reference architecture, interoperability requirements, evaluation methods, and safety requirements for Audio Large Language Models (Audio LLMs) and auditory Artificial Intelligence (AI) systems. It applies to AI systems capable of auditory perception and speech output, including but not limited to Audio Large Language Models, speech assistants, embodied agents, and general auditory perception and generation models. | Draft Development |
| P3933 | Entity | Standard for Capability Requirements of Artificial Intelligence Agents for Materials Research | This standard specifies a capability framework for artificial intelligence (AI) agents that are designed for materials research workflows. The framework comprises three domains：
(1) Fundamental capabilities help to ensure the basic operational competence required for materials-related research activities. These capabilities address data management, task management, and human–agent interaction. (2) Scientific research capabilities address essential computational, analytical, and experimental functions of materials science. These capabilities cover tasks such as materials simulation, property characterization, and experimental–computational integration. AI agents are required to explicitly state their applicability domain and data constraints. This supports transparency and controllability in materials research scenarios. (3) Service capabilities support the deployment and effective use of AI agents and enable collaboration with human experts. These capabilities address deployment, compatibility with relevant software environments, data formats, and deployment infrastructures, as well as monitoring of engineering maturity and operational reliability in practical applications. The standard also specifies agent competencies related to interoperability with external tools, data sources, and computational platforms used in materials research workflows, as well as algorithmic integration, advanced perception, and effective collaboration with human experts. Requirements for underlying model training strategies, data scale, or feature representation methods are beyond the scope of this standard. |
Draft Development |
| P3931 | Entity | Standard for Agent Description, Discovery, and Registry (ADDR) Interoperability | This standard specifies a description model, registry mechanisms, and discovery protocols for intelligent agents operating across platforms, organizations, and ecosystems. It defines an agent metadata schema—a machine-readable model for describing agent capabilities, input/output types, semantic and syntactic bindings, version and compatibility policies, a runtime environment, resource constraints, and service bindings. It also specifies registry and lifecycle interfaces, including Application Programming Interfaces (APIs) and data models for agent registration, update, and revocation, together with digital signatures, timestamps, and integrity verification to help ensure traceable and verifiable lifecycle management within interoperable registries.
The standard further defines discovery and matching protocols—methods for querying, filtering, ranking, and negotiating agent capabilities based on semantic tags, performance metrics, and version compatibility, as well as support for fallback and degradation strategies. It specifies registry aggregation and consistency—mechanisms for federated and decentralized registry synchronization, caching, conflict resolution, access control, and rate-limiting (ledger-agnostic design). It also defines validation and security mechanisms to detect and mitigate risks arising from invalid, misleading, or malicious agent descriptions, as well as conformance and interoperability testing—including test suites, compliance assertions, and plugfest procedures—for evaluating cross-platform compatibility. Identity frameworks, trust scoring systems, collaborative governance policies, or domain-specific business semantics are beyond the scope of this standard. |
Draft Development |
| P7804 | Individual | Recommended Practice for Integrated Agentic Food Waste Management Process In Schools | This recommended practice outlines procedures for applying agentic Artificial Intelligence (AI) tools and integrated logistics to help schools and their communities reduce food waste across the supply chain. | Draft Development |
| P3952 | Entity | Standard for Technical Requirements for Large-Scale Models in International Logistics Supply Chain | This standard specifies a technical framework for the application of large-scale models in the international logistics supply chain. It covers related data, models, and application services as well as interaction and collaboration layers. For these models, the standard defines business processes and functional, performance, security, and trustworthiness requirements as well as corresponding testing methods.
The standard applies to the design, development, deployment, application, operation, and management of these models. The standard provides guidance to related enterprises, technical service providers, and organizations within the international logistics supply chain. |
Draft Development |
| P3942 | Entity | Standard for Smart Power Data Query Model and System | This standard specifies the overall architecture, functional requirements, performance indicators, interface specifications, security requirements and testing evaluation for intelligent systems querying power consumption and generation data. | Draft Development |
| P3998 | Individual | Standard for Time Series Applied to Artificial Intelligence Systems—Part 1: Definition and Format of Temporal Data Records | This standard defines fundamentals, requirements, and methodologies for time series data produced internally by Artificial Intelligence Systems (AIS) across their lifecycle. It establishes a unified temporal reference framework specifying formats for internal temporal data records. This standard excludes time series external to AIS, synthetic time series, and derived data computed from AIS internal time series. | Draft Development |
| P3971 | Entity | Standard for Interoperability Specification for Scientific Intelligent Agent Systems | This standard specifies interoperability requirements for scientific intelligent agent systems. The standard encompasses a standardized description of interoperability and fundamental components of the execution framework. This standard applies to providers, operators, and users of scientific intelligent agent systems (including artificial intelligence agents and intelligent wet-lab equipment), while also supporting technological advancements based upon such systems. | Draft Development |
| P4502 | Individual | Standard for a Physical Artificial Intelligence (AI) Framework for Smart Mandibular Advancement Devices (Smart MADs) | This standard specifies the architectural framework and data interface requirements for Physical Artificial Intelligence (AI)-enabled Smart Mandibular Advancement Devices (Smart MADs), designed to perform autonomous, closed-loop therapeutic adjustments for sleep apnea and related breathing disorders. The scope encompasses the definition of logical structures and data communication protocols among the device’s core functional layers:
• the perception layer that is responsible for real-time physiological sensing and pre-processing, Specifically, this document standardizes the internal data flows and command formats between modules to help ensure interoperability across heterogeneous sensors and algorithms, while establishing mandatory control limits and fail-safe mechanisms to protect patient safety during automated physical interventions. This standard does not include clinical diagnostic criteria for specific medical conditions or the validation of the clinical efficacy of individual therapeutic algorithms. |
Draft Development |
| P4501 | Individual | Standard for Framework and Requirements of Physical Artificial Intelligence in Manufacturing | This standard establishes a foundational framework for Manufacturing Physical Artificial Intelligence (Manufacturing AX) systems operating at the interface of digital Artificial Intelligence (AI) and physical industrial environments. The scope is limited to providing a consistent basis for the entire lifecycle of Physical AI implementation through the following:
– General taxonomy and vocabulary establish a standardized terminology for Physical AI components, including sensing, cognition, decision-making, and actuation within manufacturing contexts. |
Draft Development |
| P3123 | Individual | Standard for Artificial Intelligence and Machine Learning (AI/ML) Terminology | The standard defines specific terminology utilized in artificial intelligence and machine learning (AI/ML). The standard provides clear definition for relevant terms in AI/ML. | Draft Development |
| P3488 | Entity | Standard for Technical Requirements of Artificial Intelligence-Based Data Sample Labeling and Management for Transmission and Distribution Lines | This standard specifies a framework for Artificial Intelligence (AI) based labeling and management of samples applicable to image data of transmission and distribution lines. The framework applies to the power industry, especially to organizations that wish to optimize the monitoring, maintenance, and fault diagnosis of transmission and distribution lines using AI. The standard provides a description of process and technical requirements for AI-based sample labeling and management, including sample labeling objectives, sample labeling requirements, sample labeling methods, and sample management.
The framework model employs AI algorithms to automate the process of sample labeling, helping to ensure high-quality data while also using an intelligent permission management system to maintain data security. The model not only streamlines the task of sample labeling but also enables real-time monitoring and comprehensive analysis of transmission and distribution lines, offering robust decision-making support for power system operations and maintenance. Furthermore, the framework offers guidance on data analysis, helping power companies extract valuable information from large amounts of data for line optimization and risk prevention. |
SA Ballot: Pre-Ballot |
| P4154 | Entity | Standard for Application Programming Interfaces for Cross-Platform Artificial Intelligence Model Deployment on Edge Devices | This standard specifies a series of interfaces for the deployment of Artificial Intelligence (AI) models on edge devices. The standard defines an edge model input interface, an edge model description interface, and edge execution interface, and edge output interface. The standard applies to cross-platform AI model deployment on edge devices. | Draft Development |
| P4152 | Entity | Standard for Artificial Intelligence Data Quality Evaluation for Mineral Exploration | This standard provides a unified and quantifiable framework for evaluating the quality of mineral exploration datasets that are used for Artificial Intelligence (AI). The standard specifies: (1) Data classification as a multi-dimensional classification system for mineral exploration datasets based on business scenarios, data sources, modalities, security levels, and model training stages; (2) Quality evaluation metrics as a comprehensive metric system that comprises seven core dimensions: diversity, factuality, security, timeliness, professional depth, completeness, and scenario applicability; (3) Evaluation methods that comprise both direct evaluation mechanisms and indirect evaluation mechanisms which are supported by various statistical sampling strategies; and (4) Evaluation procedures that include preparation, execution, analysis, and summary. |
Draft Development |
| P3376 | Entity | Recommended Practice for Evaluating Artificial Intelligence Generated Content | This recommended practice specifies tools and indexes for evaluating content generated by artificial intelligence technology. | RevCom Agenda(22 Sep 2026) |
| P3374 | Entity | Recommended Practice for the Framework and Evaluation Methods of Audio-Driven Portraits Based on Artificial Intelligence | This recommended practice specifies framework and evaluation methods of audio-driven portraits based on artificial intelligence. In this recommended practice, the framework includes functional modules and input/output interfaces that implement facial and body movements that drive portraits based on audio. The input interface includes the size and resolution of portraits. The output interface provides a mechanism for exporting generated videos. The framework includes modules for generating facial expressions, lip shapes, body movements, and super-resolution. This recommended practice also specifies evaluation methods based on audio-driven portrait models. | Draft Development |
| P7018 | Individual | Standard for Security and Trustworthiness Requirements in Generative Pretrained Artificial Intelligence (AI) Models | This standard establishes a comprehensive framework for mitigating security risks, privacy leaking in the development, deployment, and use of generative pretrained AI models. The framework encompasses requirements concerning security risks, functional and non-functional requirements, and evaluation metrics such as transparency, interpretability, and explainability. | Draft Development |
