|
Work item:
|
X.sr-AIec
|
|
Subject/title:
|
Security requirements for AI-Enhanced collaboration in cloud computing infrastructure
|
|
Status:
|
Under study
|
|
Approval process:
|
TAP
|
|
Type of work item:
|
Recommendation
|
|
Version:
|
New
|
|
Equivalent number:
|
-
|
|
Timing:
|
2028-03 (Medium priority)
|
|
Liaison:
|
ITU-T SG13, ITU-T SG20, ITU-T SG21, ISO/IEC JTC1/SC27
|
|
Supporting members:
|
State Grid Corporation of China，CAICT, China Automotive Engineering Research Institute Co., Ltd.
|
|
Summary:
|
Cloud computing infrastructures increasingly adopt AI-enhanced collaborative architectures that distribute and coordinate AI capabilities across cloud, edge, and device layers. These collaborative architectures enable real-time monitoring, distributed analytics, and intelligent decision-making for critical infrastructure applications including power grids, smart city services, and transportation networks. However, the complex interactions involving data exchange, model distribution, collaborative training, and distributed inference across heterogeneous environments introduce unique security challenges that extend beyond traditional AI security concerns and general cloud security measures.
This Recommendation identifies collaboration-specific security threats arising from four key characteristics of AI-enhanced collaboration in cloud computing infrastructure: devices deployed in physically accessible environments, stringent real-time operational requirements, data protection constraints, and high task dynamics. These characteristics create distinct threat vectors including device model theft and forced model downgrade, vulnerabilities introduced through model compression and conversion, gradient poisoning and inversion in federated learning, and fine-tuned backdoors in dynamic model updates. Unlike standalone AI systems where security breaches primarily affect model accuracy, these threats can cascade through the layered architecture, potentially causing incorrect control decisions, service disruptions, or safety hazards in critical infrastructure operations.
To mitigate these collaboration-induced threats, this Recommendation specifies security requirements for different collaboration types. The security requirements encompass confidentiality and isolation during distributed training workflows, model signing and integrity verification across deployment pipelines, and protection of inference data and runtime anomaly detection. By providing a structured framework that maps collaboration-specific threats to targeted security requirements, this Recommendation enables secure and reliable operation of AI-enhanced collaboration in cloud computing infrastructure.
|
|
Comment:
|
-
|
|
Reference(s):
|
|
|
Historic references:
|
|
Contact(s):
|
|
| ITU-T A.5 justification(s): |
|
|
|
|
First registration in the WP:
2025-12-11 12:46:20
|
|
Last update:
2026-01-21 10:04:52
|
|
