The Ethereum Foundation has launched zkAPI on Ethereum mainnet, a system that lets people pay for AI models and other online services from a crypto deposit without the provider learning who is paying.
It was built with the Open Anonymity Project, from a design that Vitalik Buterin and Davide Crapis, who leads the foundation’s decentralised AI team, published in February. The protocol is marked experimental.
The mechanics are simple enough. A user deposits ETH or USDC into a vault contract. Their software then generates a zero-knowledge proof that a request is backed by unspent funds, without revealing which deposit they came from.
The zkAPI server checks the proof and issues a temporary API key with a spending limit, which the user takes straight to the AI provider. A nullifier, in effect a serial number, stops the same funds being spent twice. The client speaks the standard OpenAI and Ollama interfaces locally, so existing apps can plug in.
The foundation lists AI chat and agents as the main use, alongside blockchain queries, image generation, VPN bandwidth and payments between machines.
It is a tidy answer to a real problem. Paying for an AI service usually means a card and an account, which ties every prompt to a name.
The limits are set out plainly in the launch material. zkAPI does not hide network traffic, so requests from a stable IP address can be linked, and users are pointed to Tor.
More awkward still, the model reads every prompt, and personal details, writing style or conversation history can re-link a session to a person. Anonymity at the payment layer is only as strong as what the user chooses to type.
