|
Work item:
|
X.SG-AIAD
|
|
Subject/title:
|
Security guidelines for Artificial Intelligence agent data
|
|
Status:
|
Under study
|
|
Approval process:
|
TAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-Q2 (Medium priority)
|
|
Liaison:
|
ITU-T SG21, ISO/IEC JTC 1/SC 27, SC 42
|
|
Supporting members:
|
China Unicom, China Telecom
|
|
Summary:
|
Nowadays, artificial intelligence (AI) is undergoing a paradigm evolution from generative AI (GenAI) to AI agents. Compared with passive generative AI, the AI agent is a goal-driven dynamic process that continuously interacts with the environment.
AI Agent data is one of the core elements that support agents in achieving autonomous execution. Agent data is the full-chain data of task execution and environment interaction, mainly including user interaction data, agent group collaboration data, external environment interaction data, model-related Data, memory Data, task & execution Data, system data and log data, providing a basis for continuous learning and iterative optimization for the AI agent.
Based on the architecture of AI agent (refer to ITU-T F.748.46), the agent data running through the entire processing loop of perception - cognition- planning - action - memory - interaction with the user & external environment & agent groups. The document list the essential data processing activities, processing data objects at each stage and the corresponding AI data lifecycle stage to help identify data threats..
Agent data is the cornerstone for the execution of AI agents. But, the autonomy of AI agents introduces complex data security threats and risks. These risks are deeply rooted in the AI agent data processing loop. It is necessary to conduct systematic threat identification and prevention from the perspective of AI agent data.
This document analyzes the AI agent data assets and AI agent data processing activities. Then it provides the security threats in the data processing loop and provides corresponding security requirements and security guidelines for protecting AI agents data. It can address the threats and risks cannot be fully covered by the security mechanisms of AI agent and multi-agent system. It is a supplement to existing AI agent security programs.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2026-04-02 14:18:56
|
|
Last update:
2026-06-10 16:09:59
|
