Evidence · DiggingBeagle record
A2A Failures: multi-principal coordination and verification study
Salesforce-led paper with 2,500+ A2A runs across 488 scenarios and three evaluation layers.
- Source role
- primary disclosure
Evidence record
Salesforce-led paper with 2,500+ A2A runs across 488 scenarios and three evaluation layers.
Claim-level citations (1)
- supportsA2A agents accepted peer claims while aggregate task results hid principal failures: In more than 2,500 controlled A2A runs across 488 scenarios, the Salesforce-led study reports that agents did not independently verify counterparty factual claims and that aggregate task scores could hide individual-principal failures.
Abstract; Sections 2.2, 3 and 4: three-layer outcome and peer verification