Circle Research has proposed Anduin, a reputation-rating system for AI agent service providers that ranks them using on-chain payment relationships to help prevent Sybil attacks. According to Foresight News, the researchers said existing ratings and usage metrics are easy to manipulate, with an attacker able to raise a fake provider’s rating to the market median for about $0.0027.
Anduin requires attackers to make genuine payments to independent providers already used by the target AI agent to improve a fake provider’s ranking; transferring funds between wallets does not build reputation. Tests used 66.1 million x402 payments totaling $6.1 million across 16 networks from March 23 to August 26, 2026. Anduin’s recommendation accuracy was close to that of the strongest traditional approach tested, collaborative filtering, while the cost of manipulation was about 23 times higher in the attack scenarios examined.
Circle said Anduin does not require changes to the x402 protocol or ERC standards, nor the issuance of a new token. The system remains experimental.