Researchers propose secure AI agent networks that cut energy use by 59%

Researchers Yuanyan Song, Kezhi Wang, and Xinmian Xu propose a secure wireless network for AI agents (agentic AI). Their system uses a supervisor to assign reasoning tasks while idle agents act as jammers for security. They developed two resource allocation schemes, ASC and LAW, which reduce total network energy consumption by up to 59.1% compared to benchmarks while maintaining reasoning accuracy, validated using the Qwen model.