Ethereum Now Lets You Pay for AI Without Revealing Who You Are
The Ethereum Foundation and the Open Anonymity Project launched zkAPI on Ethereum mainnet on October 1, enabling users to prepay in assets like USDC and obtain short-lived, capped API keys via zero-knowledge proofs. The design separates payment from identity so providers do not see both who paid and the content of a user's AI queries.
Why It Matters
By splitting billing from request content, zkAPI aims to give users privacy when querying remote AI models—a capability that responds to recent legal and privacy concerns about stored AI logs. If adopted, the approach could be reused for other metered services that require private payments and usage accounting.
Key Facts
- Launch date: October 1 (mainnet launch)
- Creators/design origin: Ethereum Foundation and the Open Anonymity Project; follows a February 2026 Ethereum Research proposal by Davide Crapis and Vitalik Buterin
- Payment method: Users prepay into a vault contract in ETH, USDC, etc.; balance becomes a private note
- Privacy mechanism: Client-side zero-knowledge proofs verify funds and non-reuse without revealing prompt contents or payer identity
- API keys: Provider returns capped, short-lived API keys stored only in device memory; keys capped at a dollar amount and expire quickly
The Ethereum Foundation, working with the Open Anonymity Project, has deployed zkAPI to the Ethereum mainnet. The system lets users deposit credits into a vault contract in a single on-chain transaction; from that point the balance is represented as a private note intended to be spendable only by the depositor. When a user wants to make an AI request, software on their device generates a zero-knowledge proof that demonstrates they hold sufficient funds and have not already spent them, without disclosing the underlying details.
A server operated by an AI provider verifies that proof and issues a temporary API key back to the user's device. That key is constrained by a monetary cap and lives only in volatile memory on the device. The user’s prompts are then sent to the AI provider authenticated only by the short-lived key; the provider sees the request content but not the billing information, while the payment layer sees the spending but not the prompt, so neither party alone can link identity to query content.
The project repository calls the protocol experimental, and the public documentation points to OpenRouter as the AI access layer that the short-lived keys are intended to work with. The team also published OA Chat, a browser-based private chatbot that requires no installation, and made the code available on GitHub. The documentation describes how apps that use the OpenAI API format can route requests locally to integrate with the system.
The zkAPI concept builds on prior research and proposals from Ethereum contributors. The current implementation traces back to a February 2026 Ethereum Research post authored by Davide Crapis, who leads the Foundation’s dAI team, and Vitalik Buterin. The post follows earlier Ethereum work on machine-to-machine coordination and payments, including an August 2025 proposal for an ERC-8004 standard to help autonomous agents find, verify and transact with one another. The Foundation frames zkAPI as complementing other privacy approaches—such as running local models—by covering cases where prompts still go to third-party servers.
The launch comes amid heightened privacy concerns in AI: a May 2025 U.S. court order required OpenAI to preserve output logs, including chats users had deleted, in a copyright suit. The zkAPI repository notes the protocol is experimental and does not list a formal audit, and the project documentation suggests the same plumbing could be extended to other metered services that accept proofs and settle signed usage receipts.
Keep Reading
Furious debate about THORChain vs NEAR shows idealism has limits
‘Euro stablecoin isn’t enough’: EU issuers make case for USD tokens
Bitcoin briefly hits $87K as weak US jobs data sends bond yields lower
