LLM Agent Communication Protocol (LACP) Requires Urgent Standardization: A Telecom-Inspired Protocol is Necessary

Xin Li, Mengbing Liu, Chau Yuen

Keywords: Multi-agent systems, Communication protocol, Language agents, Protocol standardization, Layered communication architecture, LLM infrastructure

TL;DR: We argue that LLM-agent ecosystems urgently require a unified, secure, and layered communication protocol—modeled after telecom standards—to overcome fragmentation and enable scalable, trustworthy multi-agent AI.

Abstract:

This position paper argues that the LLM agent field must urgently adopt a unified, telecom-inspired communication protocol, exemplified by our proposed LLM-Agent Communication Protocol (LACP), to overcome critical deficiencies in current ad-hoc approaches that threaten safety, interoperability, and scientific progress. The prevailing landscape of fragmented protocols, perilously echoing early networking's ``protocol wars'', severely curtails agent collaboration and reliability. Our analysis identifies fundamental flaws including crippling interoperability gaps that lead to scientific stagnation, inherent insecurity due to security being an afterthought, and a lack of transactional integrity stemming from monolithic designs unsuited for critical operations.

Drawing direct inspiration from telecommunications' transformative standardization, which championed principles like layered abstraction, security by construction, minimal core with extensibility, and consensus-driven interoperability, we propose LACP. LACP is a principled, three-layer framework designed to ensure agents communicate with clear semantic intent, engage in reliable, verifiable transactions, and benefit from inherent security. It embodies its core tenets—minimal core, layered design, security by default, and content agnosticism—to provide a robust and adaptable communication foundation. We urge the NeurIPS community to spearhead the adoption of such a principled approach before current fragmentation becomes an irreversible impediment to trustworthy AI, particularly in high-stakes domains. This strategic shift is vital for unlocking the full scientific and societal potential of collaborative AI.

Submission Number: 652