Ethereum Foundation Unveils zkAPI for Anonymous AI Payments

The Ethereum Foundation has introduced a new protocol designed to decouple user identity from artificial intelligence consumption. Released on the Ethereum mainnet, the system allows individuals to pay for AI models and other metered application programming interfaces without exposing personal information or linking transactions to traditional credit accounts.

Modern artificial intelligence infrastructure inherently ties every digital interaction to a specific user profile. Standard API keys connect directly to registered accounts, which in turn link to identifiable payment methods. Over time, service providers can aggregate years of individual prompts and usage patterns into comprehensive behavioral dossiers. The newly launched protocol aims to disrupt this surveillance model by introducing cryptographic privacy to routine API billing.

The mechanics of anonymous AI access

To utilize the system, users begin by depositing tokens such as Ether or USD Coin into a specialized vault contract operating on the Ethereum blockchain. This contract records the deposited balance as a private note rather than a public account ledger. When a user wishes to query an AI model, software running locally on their device generates a zero-knowledge proof. This mathematical proof verifies that the API request is backed by a legitimate, funded note without revealing which specific deposit belongs to the user.

Upon receiving the zero-knowledge proof, a dedicated server validates the cryptography and issues a temporary API key equipped with a strict spending cap. The user then transmits their prompts directly to the AI model provider alongside this temporary key. Once the session concludes, the server deducts the appropriate usage fees from the user's private balance. To prevent financial abuse, the protocol publishes a cryptographic serial number known as a nullifier with every payment. Any attempt to spend the same underlying balance twice generates a duplicate nullifier, exposing the fraud attempt while keeping the user's identity entirely obscured.

The client software is engineered to integrate seamlessly with existing developer tooling. It exposes standard OpenAI and Ollama application interfaces directly on the user's local machine. Consequently, existing productivity applications, text editors, and chat clients can route traffic securely simply by pointing their configurations to localhost.

Broader applications and current limitations

While conversational chat applications and autonomous AI agents represent the primary intended use cases, developers expect the infrastructure to support a wider array of services. Potential applications include blockchain remote procedure call queries, high-bandwidth image and video generation, virtual private network bandwidth, and automated machine-to-machine transactions between disparate AI agents.

Despite these expansive use cases, the protocol comes with notable technical limitations. The system does not inherently provide network-layer anonymity. Gateway servers can potentially correlate incoming requests if they originate from a stable Internet Protocol address. For individuals seeking robust protection against network surveillance, developers recommend routing traffic through Tor, an anonymity network that relays communications across multiple servers. Furthermore, sensitive details embedded directly within prompt content, such as personal writing styles or specific conversation histories, can still lead to re-identification.

Why it matters

The launch marks a concrete step toward realizing the Ethereum Foundation's broader ambitions regarding decentralized artificial intelligence. The protocol implements "ZK API Usage Credits," a theoretical framework originally published in February by Davide Crapis and Ethereum co-founder Vitalik Buterin. Crapis heads the Foundation's dAI team, a specialized unit formed in September 2025 to position Ethereum as a secure settlement and coordination layer for autonomous agents. This team also developed ERC-8004, a foundational standard for AI agent identity that debuted on the mainnet earlier in January.

By addressing the monetization layer of machine learning, the protocol tackles a critical vulnerability in the current technological landscape. Centralized AI providers hold absolute authority over user data and billing histories, creating severe privacy risks for enterprises and individuals alike. Cryptographic payment rails offer an alternative framework where sovereignty and data minimization are foundational rather than optional.

What to watch

Developers have explicitly described the GitHub repository for the project as experimental, meaning the code will likely undergo substantial iteration and security auditing. Observers should monitor whether major AI model providers begin accepting zero-knowledge authenticated credentials natively, or if adoption remains restricted to secondary proxy gateways. Additionally, the success of the dAI team's broader roadmap will depend on how effectively these cryptographic payment tools scale alongside competing decentralized infrastructure.