Saw Circle share this officially: Circle Research published a study introducing a framework called Anduin, designed to help AI agents that are beginning to pay for things themselves answer the question, “Which service providers should I trust?” According to Circle, this is still a research project and does not represent software or a service that has launched.

The official blog post says that ratings and call counts are easy to game. The study notes that it would cost only about $0.0027 to raise an adversarial provider’s rating to the market median. Anduin instead looks at real payment relationships on-chain (based on public records left by x402 payments): to appear trustworthy to a particular agent, a provider must spend real money on services that the agent already uses. Moving funds back and forth between your own wallets doesn’t count.

Circle says the approach was tested on 66 million real x402 payments. Its recommendation accuracy was about the same as the best conventional methods available today, but the attacks they tested cost about 23 times more to carry out.

#Circle #AI代理 #x402