Trust credentials improve ai agent choice in online marketplaces
LEGIT: Credentialing Protocol for Trustworthy AI Agent Marketplaces
Artificial Intelligence
Summary
Choosing the best AI agent to complete a task online is hard because their reported performance scores may not be trustworthy or comparable. The authors created LEGIT, a system that links an AI agent’s certified performance and costs to specific tasks and conditions, creating verifiable records. Buyers can check these credentials and past reputation to pick reliable AI agents. The system also analyzes how much it might cost to fake reputation, helping keep the marketplace honest.
What this means in practice
- •For ai platform operators: Create credential systems to verify and compare AI agents’ performance and costs before buyers select them.$Commercial implications: Enables building trustworthy AI marketplaces that sell verified agent capabilities to buyers.
- •For cloud service providers: Integrate agent certification records to assist clients in choosing AI services based on verified task outcomes and costs.
Authors
Steve Drew, Jiayu Zhou
Abstract
Agentic marketplaces are emerging where AI agents with varying capabilities autonomously complete specialized tasks for buyers. A major challenge of such marketplaces is that buyers cannot easily determine which agent will perform best on their tasks. Reported benchmark scores may be difficult to verify or compare across tasks, software, and budgets. We introduce LEGIT, a credentialing protocol connecting certification, reputation, and proposed marketplace allocation. Certification binds measured quality and cost per solved task to an agent configuration, task domain, evaluation budget, and evidence through a signed record. Reputation links records of past task outcomes to the same identity, subject to the reliability of the reported feedback. Buyers and agents can verify credential records and inspect optional visual profiles. Evaluations reveal cost differences between agent configurations with similar observed task success, and show that comparisons depend on the evaluation budget. These results support binding performance measurements to the tested configuration and resource limits. A complementary analysis quantifies the deposits and fees required for reputation manipulation under a stated Sybil attack model.