The Ethereum Foundation has launched zkAPI, a privacy focused payment system designed to let users pay for AI models and other metered APIs without linking their identity to their requests.
Built with the Open Anonymity Project, zkAPI is now live on Ethereum mainnet. The system uses zero knowledge proofs to separate payment information from API usage, allowing a service provider to verify that a user has paid without learning which deposit funded the request or who made it.
KEY TAKEAWAYS
- zkAPI allows users to fund API usage with ETH, USDC and other supported credits through an Ethereum vault.
- Zero knowledge proofs verify that a user has sufficient funds without exposing the underlying deposit or payment identity.
- AI providers can receive prompts without being given the user’s billing identity.
- The system also targets blockchain RPC, image and video generation, VPN bandwidth and machine to machine payments.
- zkAPI does not provide complete anonymity because IP addresses, timing information and prompt content can still reveal information.
- The protocol is currently described as experimental in its public GitHub repository.
SEPARATING AI USAGE FROM PAYMENT IDENTITY
Traditional AI APIs generally require an account or API key connected to a payment method. That creates a persistent relationship between a customer’s billing information and the requests they make. zkAPI takes a different approach. Users first deposit funds into a vault on Ethereum. The deposit is represented as a private note rather than a conventional account balance. When the user wants to access a metered service, software running on their device generates a zero knowledge proof demonstrating that the request is backed by an eligible, unused balance.
The proof does not reveal which deposit belongs to the user. Instead, the zkAPI server verifies the proof and provides a temporary API key with a spending limit. The user’s request can then be sent to the AI provider without passing the user’s payment identity along with it. The system also uses nullifiers to prevent the same balance from being spent twice. A nullifier acts as a cryptographic marker for a spend without exposing the private note behind it. The result is a separation between the payment and inference layers. The payment server can verify spending authorization, while the AI provider handles the request itself without receiving the corresponding billing identity.
FROM ETHEREUM RESEARCH TO A WORKING MAINNET SYSTEM
The technology builds on a proposal published by Ethereum Foundation dAI lead Davide Crapis and Ethereum co-founder Vitalik Buterin on Ethereum Research in February. Their proposal identified a privacy problem with existing API payment models. Linking every request to an account can allow providers to build long term profiles of users, while making individual onchain payments for every request can be costly, inefficient and publicly traceable. The proposed solution was to allow users to deposit funds once and authorize multiple API calls privately.
zkAPI turns that concept into a functioning implementation. The project’s public repository includes the client, browser SDK, contracts and supporting infrastructure required for the protocol. It also provides an OpenAI compatible local client, allowing compatible applications to interact with the system without requiring a completely different interface. The launch also fits into the Ethereum Foundation’s broader work around AI infrastructure. The foundation’s dAI team has been exploring Ethereum’s potential role in AI related payments, coordination and agent infrastructure.
PRIVACY HAS LIMITS
zkAPI does not make an AI user’s entire activity anonymous. The Ethereum Foundation acknowledges that network information can still expose users. A stable IP address, request timing and traffic patterns could potentially allow sessions to be correlated. Prompt content creates another potential source of identification. Repeated personal information, writing patterns or unique project details could allow an AI provider to associate otherwise separate sessions.
That distinction means zkAPI primarily addresses the link between payment identity and API usage, rather than providing complete network or content privacy. The project’s GitHub repository also describes the protocol as experimental. Its current implementation uses cryptographic components including Groth16 proofs over BN254, Poseidon hashing and a 32 level Merkle tree.
BEYOND AI PAYMENTS
Although private AI inference is the first major application, the same payment model can be applied to other metered services. The Ethereum Foundation lists blockchain RPC queries, image and video generation, VPN bandwidth and payments between autonomous machines as potential applications. This could become particularly relevant as AI agents increasingly interact with paid digital services without direct human involvement.
For now, zkAPI represents an attempt to solve a specific privacy problem: allowing a user to pay for digital services without making the payment relationship the permanent identity attached to every request.
CONCLUSION
The zkAPI launch moves Ethereum’s zero knowledge research into a live system for private API payments. By separating payment authorization from service requests, the technology gives users a way to pay for AI services without automatically attaching their billing identity to their prompts.
Its protections remain limited by network metadata and information contained in the requests themselves, and the protocol is still experimental. Nevertheless, the mainnet release provides a working implementation of the private API payment design originally proposed by Crapis and Buterin earlier this year.
