The Ethereum Foundation has launched zkAPI on the main Ethereum network. It lets you pay for AI model requests so that nobody can tie you to your prompts. If you have ever asked a chatbot about your health or your money, this concerns you.
What exactly launched?
zkAPI was built with the Open Anonymity Project, an open initiative on AI privacy. The system is already usable, and the code is public on GitHub. It ships together with OA Chat, a private browser chatbot that needs no installation.
The idea is not new. Ethereum co-founder Vitalik Buterin and dAI team lead Davide Crapis published it on the Ethereum Research forum in February. Now it is a working product.
What is wrong with regular AI billing?
To see why this matters, recall how services work today. Your API key points to an account, the account to a payment method, and every prompt lands in one shared history. According to the project's authors, a provider can stitch years of usage into a single profile.
"Prompts are personal. People ask AI models about their health, their finances, their doubts. Under the current model, using AI means handing a running transcript of your thinking to whoever holds the billing relationship."
- Ethereum Foundation, from the zkAPI launch blog post of October 1, 2026
The concern is real. In May 2025 a US federal court ordered OpenAI to preserve its output logs, including chats users had deleted, in a lawsuit brought by The New York Times and other publishers.
How does it work step by step?
The mechanics boil down to a few simple actions, even though the math underneath is heavy.
- Deposit: you put ETH or USDC into a vault contract on Ethereum with one ordinary transaction, and your balance becomes a private note.
- When you need to send a request, software on your device builds a zero-knowledge proof. It is a mathematical receipt showing that the funds exist and are unspent, without revealing which deposit is yours.
- A server checks the proof and issues a temporary API key with a dollar cap. The key lives only in your device's memory.
- The request goes to the provider with that key and nothing else. No name, no payment details.
- When the key expires, the provider issues a signed usage receipt, and you are charged for actual usage, not the full cap.
A serial number called a nullifier guards against double spending. Trying to spend one balance twice produces a duplicate number, so the system sees the attempt and learns nothing else about you.
What does it work with, and where are the limits?
The client exposes the standard OpenAI and Ollama interfaces on your own machine. That means familiar apps, editors and chat clients connect by simply pointing at localhost. Per the documentation, short-lived keys work with OpenRouter, a service that gives access to hundreds of models through one connection.
The first use case is AI chat and agents. Planned extras include blockchain node queries, image and video generation, VPN bandwidth and machine-to-machine payments. For a provider to integrate directly, it has to accept a proof instead of a key and settle with signed usage receipts.
The authors list the limits themselves. zkAPI does not provide network anonymity, so the gateway can correlate requests from a stable IP, and users who want stronger protection are advised to route traffic through Tor. Sessions can also be re-linked through prompt content, such as personal details or writing style. The repository labels the protocol experimental, and no formal audit is listed.
What it means for Ethereum
This is part of the network's wider bet on AI. The Foundation formed its dAI team in September 2025 to make Ethereum a settlement and coordination layer for AI. The same team developed ERC-8004, an identity standard for AI agents that went live on mainnet in January.
For an ordinary user the takeaway is simple: payment privacy becomes a feature in its own right, not a side effect. For now this is an experiment without an audit, so large deposits are premature. Still, the direction shows crypto has found a job for AI that a bank card does not solve.




Comments
Your email address will not be published. Required fields are marked *