Ethereum Foundation Deploys zkAPI for Anonymous AI Payments

The Ethereum Foundation switched on zkAPI in the mainnet environment on October 1, 2026, a protocol that lets people pay for AI services and metered API access without tying the request to the wallet that funded it. According to Blockonomi, the system runs on vault deposits of ETH or USDC, with zero-knowledge proofs authorizing each spend.

The work was done with the Open Anonymity Project, which builds open-source privacy tools for AI. The Foundation argued that conventional API credentials fold identity into every request, because an API key points to an account, the account to a payment instrument, and each prompt is then filed against both. The design was laid out in a February proposal by Vitalik Buterin and Davide Crapis, who directs the Foundation’s dAI initiative. Cryptopolitan reported that the dAI programme launched in September 2025 and that the same team shipped ERC-8004, an AI agent identity standard, to mainnet in January.

Key facts

• zkAPI went live on Ethereum mainnet on October 1, 2026, developed with the Open Anonymity Project.
• Users fund an Ethereum vault contract with ETH or USDC, which generates private notes with no publicly visible owner.
• Spending is authorized by a zero-knowledge proof generated on the user’s own machine; each spend produces a unique nullifier that flags double-spending attempts.
• Vitalik Buterin and Davide Crapis introduced the concept in a February proposal on the Ethereum Research platform.
• Cryptopolitan reported the Open Anonymity Project’s Ken Liu described the system as a generalization of OA unlinkable inference that also abstracts payments away.

How the payment flow works

A user makes one standardized transaction, depositing ETH or USDC into a vault contract on Ethereum. The deposit becomes digital money the user can spend, but the vault records no public link between the funds and the depositor. When a request goes out, the local client builds a zero-knowledge proof confirming the account holds enough balance without disclosing which deposit it comes from. A zkAPI gateway checks that proof and issues temporary API credentials with preset spending caps; the user then talks directly to the AI provider through those short-lived keys.

Payment runs separately from the service request itself. Every spend emits a nullifier, a serial number that prevents the same funds from being used twice. Duplicate nullifiers are caught without exposing who is spending. On the client side, the software mimics the standard OpenAI and Ollama API protocols on localhost, so existing chat clients, editors and applications can connect without being rewritten. The Foundation lists blockchain data retrieval, multimedia generation, VPN access payments and autonomous agent-to-agent transactions as further uses. The GitHub repository labels the protocol experimental.

Where the anonymity stops

Two limits are explicitly acknowledged. The protocol does not mask network-level identifiers, so gateway operators could correlate repeat requests coming from a consistent location; the Foundation suggests routing through privacy networks such as Tor where stronger anonymity is needed. Prompt content is the second gap. Distinctive phrasing, references to personal details or an ongoing thread of conversation can still connect separate interactions back to one person.

Why it matters

Paid inference and metered APIs are now the default way AI products are billed, and that billing model has generally required an account, a card and a persistent usage record. zkAPI targets that record specifically, letting a payment credential exist independently of the service account. For developers, the appeal is that the privacy layer sits between application and provider rather than requiring a rebuild. Whether it changes behaviour depends on integration work, demand and how the system performs under heavier transaction volume.

What to watch

Watch whether independent developers adopt the localhost OpenAI and Ollama shims and how the gateway ecosystem handles request correlation. The repository’s experimental label means the design can still change, and Cryptopolitan noted that network-wide work continues separately, with Glamsterdam’s execution-layer parallel processing expected in October.

Nothing here is financial advice, and crypto and AI-service markets are volatile and uncertain.

Disclaimer: This article is for information only and is not investment, financial or trading advice. Cryptocurrency prices are highly volatile. Always do your own research.

Sources: Blockonomi, Cryptopolitan