The Ethereum Foundation and the Open Anonymity Project activated zkAPI on Ethereum mainnet on October 1, 2026, deploying a smart contract vault that lets users fund API calls without linking payments to a specific deposit or identity.

The system targets a narrow but growing demand: developers and end users who want to consume AI and other API services without exposing payment provenance to providers or observers. zkAPI routes authorization through zero-knowledge proofs rather than traceable wallet signatures, an approach the partners frame as addressing both commercial privacy and surveillance risk.

How does zkAPI sever the payment trail?

A user deposits ETH or USDC into the zkAPI vault contract on Ethereum mainnet. The vault converts those balances into USDC credits at the contract level. When the user calls an external service, such as a large language model provider, the system generates a cryptographic proof authorizing the payment.

The service receives its funds. The user receives the API response. On-chain analysis cannot connect the payment to the originating deposit, because the proof never exposes which credit was consumed.

What cryptographic primitives power the system?

zkAPI combines three components, each handling a discrete step in the unlinkability chain:

• Groth16 proofs, a zero-knowledge scheme that lets a prover demonstrate a statement is true without revealing the underlying data.
• Baby-JubJub commitments, a binding scheme that locks in a deposit value so it can be referenced later without exposure.
• A 32-level Merkle tree, a data structure that aggregates many commitments into a single root hash, allowing membership proofs without identifying a specific leaf.

Together, the three primitives split the authorization workflow: Groth16 proves authorization, Baby-JubJub commits to the deposit, and the Merkle tree establishes the commitment within an aggregated set without revealing its position.

Where does custody sit?

Funds remain in contracts controlled on Ethereum itself, not on a company-operated server. A depositor can withdraw directly from the vault, independent of any operator's cooperation. The structure removes single points of failure for both censorship and insolvency.

Withdrawal independence carries direct censorship-resistance implications. Any party seeking to block a user from funding API access would need to censor Ethereum itself, not a single gateway operator.

Where did the design originate?

The architecture traces to a February 11, 2026 research post by Vitalik Buterin and Davide Crapis. That paper laid out the case for unlinkable payment rails for AI and other metered API consumption.

The timeline from the February research post to the October mainnet deployment came in under eight months.

Who is funding the project?

The Open Anonymity Project had secured more than $100,000 in grants from the Ethereum Foundation and the Human Rights Foundation as of May 2026. The Human Rights Foundation's involvement positions financial privacy as a civil-liberties matter, not solely a commercial one.

What changes in the AI payments stack?

zkAPI enters a market dominated by centralized AI gateways, where payment flows typically expose wallet addresses, billing details, and usage patterns to providers and on-chain observers. Adoption at scale could pressure centralized AI platforms to integrate similar unlinkability primitives, or to accept zkAPI credits as a payment rail alongside traditional processors.

The next milestone worth tracking: whether AI providers begin accepting zkAPI credits directly at the API edge, rather than forcing users to revert to traceable payment methods.