Ethereum Foundation Launches zkAPI for Private AI Payments
The Ethereum Foundation and Open Anonymity Project have made zkAPI available on Ethereum mainnet, aiming to hide the link between payments for AI services and the requests users submit.
Announced in a blog post by the Ethereum Foundation, zkAPI addresses the risk of being tracked through payment information when using AI. In today’s systems, an API key can link a user’s account and payment method to the questions they send to an AI service. The new tool aims to prevent this link from being established when users pay for a service.
The system is based on a design published in February by Vitalik Buterin and Davide Crapis from the Ethereum Foundation’s dAI team. Developed with the open-source privacy project Open Anonymity Project, the tool makes this approach usable on Ethereum.
What changes when you pay for AI?
First, users deposit tokens such as ETH or USDC into a vault contract on Ethereum and create a private balance. Their device then uses a zero-knowledge proof to show that the balance is sufficient to cover the request, without revealing which deposit transaction it is linked to.
Once the server verifies the proof, the user receives a temporary API key with a spending limit. The user sends questions directly to the AI service provider using this key. When the key expires, the usage fee is deducted from the private balance. This lets the service provider verify payment authorization while keeping secret which deposit funded the usage.
Works with existing AI tools
The zkAPI client supports OpenAI and Ollama API formats. Existing apps and chat tools can be run through the system by routing them to a local connection address on the user’s computer.
Although its initial use cases are AI chats and agents, the approach was designed to work with other services that charge based on usage. Potential applications include blockchain queries, image and video generation, VPN bandwidth, and payments between AI agents.
Where are the limits of privacy protection?
The protection focuses on the link between payment and request; it does not hide internet traffic. Requests sent from a fixed IP address can be linked to one another, while personal information entered into an AI service, writing style, and conversation history can also make it possible to recognize a user across different sessions.
The project’s GitHub repository states that zkAPI is still experimental.