Trust, policy, and teamwork: Missing layers in standards for multi‐agent system interoperability

Multi‐agent system (MAS) interoperability falls short in three crucial areas: trust, policy, and teamwork . First, we argue that the requirement for trust goes far beyond the basic requirements of authentication. Importantly, accurate trust calibration requires access to robust, verifiable, continuously updated evidence based on experience. Second, we describe the value of digital policy services . Complementing internal mechanisms to ensure agent safety, policy services regulate agents externally —that is, without requiring their consent or cooperation. Third, in contrast to many common orchestration approaches for multi‐agent cooperation, agent teamwork support mechanisms encourage “combine and succeed” rather than “divide and conquer” strategies. To illustrate such approaches in practice, we present three specific examples: (1) blockchain technology adapted to MAS as a trust monitoring and orchestration substrate, (2) sophisticated digital policy services (KAoS), and (3) a KAoS‐enabled agent teamwork framework (Luna). We believe that greater attention to the missing interoperability layers of trust, policy, and teamwork will enable safer and more widespread deployment of MAS in consequential applications. In the spirit of the Multi‐Agent System Safety and Teamwork (MASST) initiative, we look forward to collaborating with like‐minded partners to bring effective open‐source solutions for the missing interoperability layers outlined here—and others—to full fruition.