The Ethereum Foundation (EF) has introduced a crypto-native answer to a growing privacy concern in artificial intelligence: a new system called zkAPI that lets users settle transactions for AI models and metered APIs under a complete veil of anonymity. The technology was co-developed alongside the Open Anonymity (OA) Project.

Under conventional payment flows, users have had to tie a credit card or API key to every request—an arrangement criticized as vulnerable because it exposes prompt history to infrastructure providers. According to the Ethereum Foundation, the new architecture resolves that problem by handling spending authorization locally through cryptographic notes.

Separating payments from identity

In a blog post published Thursday, the Ethereum Foundation acknowledged the deep-seated privacy flaws built into modern tech architecture. “Your API key points to an account, the account to a payment method, and every prompt you send joins the record attached to both,” the post noted.

With zkAPI, the Foundation said, payment is completely decoupled from identity. The process begins with a single standardized transaction in which a user deposits funds—such as ETH or USDC, a US dollar stablecoin—as a credit deposit into a vault contract on Ethereum. The funds become purely digital money, fully spendable by the user, yet no longer traceable to their identity or to any previous depositing activity.

Authorization then takes place entirely on the side. A zero-knowledge proof—a cryptographic technique that lets one party prove a statement is true without revealing the underlying information—is created locally, demonstrating that the transaction falls under a private note without disclosing the state. More importantly, a single ZK proof can verify both an individual transaction and an entire session, according to the EF.

The architecture also employs a distinctive serial number, known as a nullifier, to secure the spending process. In the event of any double-spending attempt, the system simply spots the duplicate nullifier without disclosing any user information.

In addition, the default APIs from OpenAI and Ollama are available locally, enabling existing AI applications, editors, and chat clients to access them via localhost.

Speaking on the new system, Ken Liu, a Stanford computer science PhD candidate who works on the Open Anonymity Project, said: “It’s like a generalization of OA unlinkable inference that also abstracts payments away.”

zkAPI could reshape privacy for AI services

The development could prove particularly relevant as AI applications increasingly rely on paid inference and metered services. Instead of requiring users to maintain accounts with individual AI providers, zkAPI creates a payment layer that could potentially allow applications to pay for usage without exposing the identity behind each request.

The distinction matters because AI interactions can contain highly sensitive information. Prompts may reveal personal details, business strategies, proprietary code, financial information, or other private data. If API usage can be linked to a persistent account or payment method, infrastructure providers may be able to build detailed records around that activity.

The Ethereum Foundation’s approach attempts to remove that link by making the payment credential independent from the user’s identity. Users can fund a private balance and authorize spending without repeatedly presenting conventional account credentials to the service provider.

The architecture could also make privacy-preserving payments easier to integrate into existing applications. Because zkAPI exposes standard interfaces for services such as OpenAI and Ollama through localhost—the address applications use to reach services running on the same machine—developers would not necessarily need to redesign their AI applications from scratch. Instead, the privacy layer can sit between the application and the underlying API service.

However, the technology is still an emerging approach, meaning its broader adoption will depend on developer integration, user demand, infrastructure support, and the system’s performance as transaction volumes increase.

Ethereum targets network scale and user experience

The announcement lands as the Ethereum Foundation pushes forward with its 2026 protocol priorities. The engineering roadmap positions the ecosystem’s immediate development on a three-fold approach: network scaling, user experience improvements, and Layer-1 fortifications.

Ethereum’s developers have already begun implementing quantum-proof cryptographic algorithms in both the execution and consensus layers, laying the foundation for complete quantum resistance by 2029. The gas limit targeted by the network in 2026 stands at more than 100 million. In November, Ethereum educator Anthony Sassano reported that hitting 180 million is the year’s ideal goal, though not the very best outcome achievable. In 2025, the Ethereum gas limit was increased to 60 million thanks to the Pectra and Fusaka upgrades.

October is also set to see the implementation of Glamsterdam on the Ethereum blockchain, viewed as one of the biggest milestones for the network since the Merge in 2022. The upgrade will be rolled out in a dual-layer manner, consisting of the “Amsterdam” update for the execution layer and the “Gloas” update for the consensus layer. On the execution side, Amsterdam will introduce Parallel Processing, which, as the name suggests, changes the execution layer to process transactions in parallel rather than sequentially—like moving past a single-lane checkout counter at a supermarket.

Source: Cryptopolitan