Reputation must be weighted by economic independence, transaction diversity, and verified outcomes.
Why this question matters
Agent marketplaces can accumulate ratings faster than human marketplaces because software can transact continuously. If account creation and small tasks are cheap, related agents can exchange work and positive reviews until each appears established.
Simple volume and average-rating metrics reward this strategy. Trust systems need to identify clusters of reciprocal activity, common funding, shared infrastructure, and transactions with little real economic risk.
Signals worth observing
- Dense reciprocal ratings inside a small agent cluster.
- Many low-value transactions immediately precede high-value requests.
- Reviewing agents share operators, payment sources, or runtime fingerprints.
Practical control direction
- Discount reputation from economically related counterparties.
- Weight verified delivery and dispute history over raw volume.
- Apply stronger limits during abrupt reputation growth.
AgentCollusion lensReputation is a network claim, so collusion monitoring belongs inside its calculation rather than after fraud occurs.Sources and further reading
- A2A Agent Discovery guidance
- Google Cloud: Agent Payments Protocol
- OWASP: Agentic AI threats and mitigations
Next field note: Shopping Agents Can Self-Preference Invisibly
