Agentic societies require social harnesses for secure coordination
Agentic Societies Need a Social Harness
Multiagent SystemsArtificial IntelligenceNetworking and Internet Architecture
Summary
When many AI agents work together for different people or groups, they often struggle to cooperate well and can be tricked by bad actors. The authors show that current communication methods between these agents are vulnerable and can cause problems. They suggest adding a new layer called a social harness to help agents communicate safely, spot bad messages, and handle problems after they happen. This would make interactions among AI agents more reliable and secure.
What this means in practice
- •For multi-agent system developers: Design communication architectures that include social harness layers to improve safety and reliability among autonomous cooperating agents.
- •For enterprise ai integrators: Implement social harnesses to prevent malicious actions and message exploitation in AI systems coordinating for different business units.
A position paper. It proposes an approach and reports no results.
Authors
Tapan Chugh, Vidushi Singh, Krish Jain, Arvind Krishnamurthy, Ratul Mahajan
Abstract
An agentic society is a collection of AI agents that coordinate autonomously across trust boundaries, on behalf of different principals whose objectives may only partially align. We show experimentally that in agentic societies even honest, competent agents often fail to reach satisfactory outcomes with existing harnesses and messaging primitives, and that faulty or malicious agents can stall collaboration, influence outcomes, and pursue other harmful goals by exploiting vulnerabilities in communication (``speech''). We argue that agentic societies need a \emph{social harness} for inter-agent interactions, in addition to each agent's \emph{personal harness}, which manages its private context and communication with its principal. We propose a layered architecture for social harnesses which (i) prevents classes of failures outright, (ii) enables agents to detect invalid messages at runtime, and (iii) supports post-facto investigation and consequences, and highlight directions for future research to realize these capabilities.