TLDR
- The Ethereum Foundation launched zkAPI on mainnet on October 1, 2026.
- The system lets users pay for AI models and other APIs without revealing their identity.
- Users deposit ETH or USDC into a vault and use zero-knowledge proofs to authorize spending.
- The design was first proposed in February by Vitalik Buterin and Davide Crapis.
- zkAPI does not hide prompt content or network metadata, so some tracking is still possible.
Ethereum took a new step toward private payments this week. The Ethereum Foundation announced that zkAPI, a privacy tool for paying for AI services, is now live on Ethereum mainnet.
🚨HUGE: Ethereum Foundation and Open Anonymity Project launch zkAPI on mainnet, letting users pay for AI WITHOUT linking their requests to the wallet that paid.
Based on a design co-authored by Vitalik Buterin, zkAPI lets users prepay with ETH or USDC and use zero-knowledge… pic.twitter.com/GbF0ffV4Ur
— Coin Bureau (@coinbureau) October 2, 2026
The system was built together with the Open Anonymity Project. That group focuses on AI privacy tools that are open source.
zkAPI lets people pay for AI models and other metered services without linking the payment to their identity. This solves a problem the Foundation described in its blog post.
“Every AI API call today carries an identity,” the post said. It explained that a person’s API key connects to an account, which connects to a payment method. Each prompt then joins a growing profile tied to that person.
How the System Works
To use zkAPI, a person first deposits tokens into a vault contract on Ethereum. They can use ETH or USDC for this step.
The vault records the deposit as a private note. This note does not show who owns it.
When the person wants to make a request, their device creates a zero-knowledge proof. This proof shows they have enough funds without revealing which deposit is theirs.
A zkAPI server checks this proof. It then issues a short-term API key with a spending limit attached.
The user sends prompts straight to the AI provider using this temporary key. The payment system handles billing separately from the content of the request.
Each transaction also creates something called a nullifier. This is a serial number that prevents someone from spending the same balance twice.
If a duplicate nullifier appears, it signals an attempted double-spend. The system flags this without exposing any other private details.
The client tool works with existing AI apps. It exposes standard OpenAI and Ollama style APIs on a person’s own device, so current apps and chat tools can connect by pointing to localhost.
What zkAPI Can and Cannot Do
The Foundation listed several uses for zkAPI beyond AI chat. These include blockchain data queries, image and video generation tools, VPN bandwidth payments, and payments between AI agents.
However, the system has real limits. zkAPI does not hide a user’s network metadata or IP address.
This means a gateway could still link requests sent from the same stable connection. People who want stronger privacy may need to route their traffic through Tor.
Prompt content itself can also create links between sessions. Personal details, writing style, or conversation history could tie separate requests back to the same person.
The project’s GitHub page describes the protocol as experimental. This suggests more testing and changes may follow.
The design behind zkAPI did not start this year. Davide Crapis, who leads the Ethereum Foundation’s dAI team, and Vitalik Buterin first published the concept in February on the Ethereum Research forum.
That team was formed in September 2025. Its goal is to build out Ethereum’s role as a settlement layer for AI systems.
The dAI team also created ERC-8004 earlier this year. That standard addresses identity for AI agents and went live on mainnet in January.
Ken Liu, who works on the Open Anonymity Project, commented on the launch Thursday. He said the system builds on existing anonymity work while also handling payments.
He added that privacy and user control should be built into more internet infrastructure going forward.
