Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States
Sarath Shekkizhar, Prathyusha Jwalapuram, Zixuan Ke and colleagues (Salesforce, with SUTD and NTU) study agent-to-agent systems in which each agent serves a different principal and holds private state, and derive a 20-mode failure taxonomy from more than 2,500 recorded runs.
Ask this paper
Setting. Agents act for different principals, each with its own environment, information and authority. Interests can be cooperative, mixed-motive or competitive, which earlier multi-agent failure studies with one orchestrator or one shared workspace did not cover.
Scale. More than 2,500 runs across several harnesses and models, covering 488 scenarios in eight families including legal, CRM, security and commerce.
Taxonomy. A^2-Failures has 20 failure modes in four groups: discovery and delegation, grounding and record consistency, verification, and conduct and fidelity.
Three consistent findings. Agents delegate to peer names that appear in the prompt instead of searching for available peers. Agents never fact-check each other unless prompted. Scoring only the final output credits the lead agent and hides coordination and verification failures.
Release. The run logs behind the taxonomy are public on Hugging Face.
Abstract
Agents are increasingly interacting with other agents that act for different principals -- people or organizations on whose behalf they work -- e.g., a buyer's personal agent negotiating with an enterprise's seller agent. In such agent-to-agent (A2A) scenarios, each agent has its own environment and holds information and authority that the other cannot see. Moreover, the interests of each agent may be aligned (cooperative), partly aligned (mixed-motive), or opposed (competitive). Existing multi-agent system failure studies assume only one orchestrator, shared workspace, or principal. We ask how A2A systems fail when these assumptions are removed. We study more than 2,500 recorded runs of A2A systems built from different agent harnesses and models, spanning 488 scenarios across eight families (legal, CRM, security, commerce, etc.). From these traces, we derive $A^2$-Failures, a taxonomy of 20 A2A failure modes in four categories: (i) discovery and delegation, (ii) grounding and record consistency, (iii) verification, and (iv) conduct and fidelity. Three observations hold across the harnesses, models, and scenarios studied: (1) Agents match tasks to peer names mentioned in the prompt rather than actively searching for available peers; (2) Agents never fact-check each other unprompted; (3) Scoring only the final output rewards the lead agent while hiding internal coordination, verification, and behavioral breakdowns. Our study makes a case for comprehensive evaluations where agents are scored individually against their own objectives in addition to task outcome. We release the run logs behind the taxonomy at https://huggingface.co/datasets/a2a-salesforce/a2-failures.