The Ethereum Foundation and the Open Anonymity Project launched on Thursday (1st) the zkAPI, a system that allows paying for the use of APIs without directly linking each session to the identity or the deposit that funded it. The infrastructure is already available on the Ethereum mainnet and uses zero-knowledge proofs to authorize service consumption.
The first focus is artificial intelligence APIs, in which an access key is normally associated with an account and a payment method. With zkAPI, that relationship is split: the system responsible for billing processes the payment, while the AI provider receives the requests, without either of them alone having the complete link between identity, payment, and usage.
The user deposits credits, such as ETH or USDC, into a contract on Ethereum. The balance is then represented by a private note, and the software on the device generates a cryptographic proof to demonstrate that there are sufficient funds for a given session, without indicating which deposit is being used.
After verifying the proof, the zkAPI server creates a temporary key with a spending limit. Requests are sent to the API provider with that key and, at the end of the session, the recorded usage is deducted from the private balance by means of a signed receipt. A single authorization can cover multiple calls.
In the architecture described by the Ethereum Foundation, the payment server knows that there is a valid payment and how much was spent in the session, but it does not receive the content of the requests nor identify the deposit used. The AI provider, in turn, continues to see prompts and responses, but does not receive the billing identity associated with the key.
Privacy does not cover content and connection
The launch does not mean that every interaction with an API becomes anonymous. The Ethereum Foundation itself highlights that the provider can still observe the content received and that network information, such as the IP address, can allow correlations between different sessions.
Content can also reveal patterns. Personal details, writing style, reused history, and documents sent can work as identifiers, allowing the provider to relate sessions that the payment layer kept separate. zkAPI, therefore, acts mainly on the relationship between billing and usage, and not on all possible forms of user identification.
Although AI inference is the first use case, the same infrastructure can be applied to services billed according to consumption, including queries to blockchain networks, image and video generation, VPNs, and automatic payments between machines or AI agents.
zkAPI implements a model previously developed by Davide Crapis, of the Ethereum Foundation, and Vitalik Buterin. In addition to the active contract on mainnet, the project has made available a local client, server, browser tools, and integration with applications compatible with APIs in the OpenAI standard.
