Answer

In a multi-agent system, multiple AI models work together — one might plan tasks, another executes them, and another checks the results. For this to work reliably, agents need to act on each other's outputs without constant human verification. This creates a form of inter-agent trust.

But trust between AI agents is fragile in ways human trust isn't. An agent has no shared history, social intuition, or common sense to fall back on. Trust must be encoded through reputation mechanisms, output verification, or constraints that limit how much one agent can influence another.

This becomes a governance challenge as systems scale. If one agent is compromised or simply wrong, and others trust it blindly, errors can cascade silently through the entire pipeline — making the design of trust boundaries one of the most important unsolved problems in multi-agent AI.