Deep Web3

SYSTEMS.
PEOPLE.
POSSIBILITIES.

Menu

Go beneath
the surface.

Anduin Proposes Payment-Graph Reputation for AI Agent Services

Originally reported by Foresight News on November 13, 2026, at 23:34 UTC+8.

Circle Research has proposed Anduin, an experimental reputation system that ranks AI Agent service providers using the on-chain record of who pays whom. Its aim is to make rankings harder to manipulate with fake providers or wallets—a form of Sybil attack.

The research argues that ratings and usage counts are easy to fabricate: an attacker could reportedly raise a fake provider’s rating to the market median for about $0.0027. Under Anduin’s proposed approach, raising that provider’s position in a particular AI Agent’s ranking would require real payments to independent providers the agent already uses. Moving funds between wallets controlled by the attacker would not itself confer reputation.

Circle Research tested the approach against 66.1 million x402 payments totaling $6.1 million across 16 networks, dated March 23 through August 26, 2026. It reported recommendation accuracy close to that of the strongest conventional method it compared, collaborative filtering. In the attack scenarios tested, manipulation cost was approximately 23 times that of collaborative filtering. Those are results from the reported tests, not evidence that every possible attack would carry the same cost.

Circle said Anduin would require no changes to the x402 protocol or ERC standards and no new token. It remained experimental at the time of the report.

Original report