Vitalik wrote a proposal on how to secretly use AI large language models

Author: Deep Tide TechFlow

Everyone around the world is talking about AI, but the discussions about crypto have quieted down quite a bit.

At the same time, ETH has been hovering around 2000 for nearly two months. Vitalik’s words and actions seem to have less impact now.

However, I recently checked his X account and found that AI’s influence on him is more than just us. Over the past month, a large portion of his posts are related to AI, even to the level of specific technical solutions.

The most noteworthy is a proposal he co-published on ethresear.ch on February 11 with Davide Crapis, the head of AI at the Ethereum Foundation, titled “ZK API Usage Credits.”

In a nutshell: Using zero-knowledge proofs to enable anonymous access to large AI models.

Currently, whether you’re using ChatGPT or calling Claude’s API, there’s only one payment method:

Register an account, link an email, and attach a credit card.

Every conversation or prompt you send, the platform knows it’s you. What you asked, when you asked, and how many times—all tied to your real identity.

Vitalik and Crapis’ proposal offers an alternative.

Users deposit a sum of money into a smart contract, for example, 100 USDC.

The contract registers this deposit on a cryptographic list on the blockchain. Then, each time you call the API, you don’t need to reveal your identity—just generate a zero-knowledge proof.

This proof can demonstrate two things to the service provider: that you are on the list and that your balance is sufficient. But the proof itself doesn’t reveal which one on the list you are.

The service provider receives the payment and can prevent abuse, but from start to finish, they don’t know who you are.

You can interpret this proposal as Vitalik’s belief that in the AI era, users shouldn’t have to give up their identities just to use an AI tool.

This proposal is still in the research stage and has a long way to go before implementation. Major AI model providers might not agree with such an approach; meanwhile, the comment section is full of rebuttals and doubts, arguing that AI model companies will always find ways to identify users.

But I believe the significance of this proposal isn’t solely whether it can be realized.

Privacy has been a decade-long focus for Vitalik. From early support for Tornado Cash to promoting zero-knowledge proofs as a core technology for Ethereum, this thread has never been broken. However, in recent years, privacy in the crypto industry has lacked a compelling story to carry it forward.

AI has filled that gap. When you talk to large models more than anyone else every day, privacy becomes a real need.

Vitalik Embraces AI

Since February, a significant portion of what Vitalik posts on X relates to AI, with such frequency that it doesn’t seem like casual chatter.

Yesterday, he posted a long message saying he recently attended a cryptography conference where people cared about privacy, open source, and anti-censorship… but had no emotional connection to blockchain.

Among that crowd, he conducted a thought experiment:

Forget “we are Ethereum community,” and start from zero—where is Ethereum most useful?

His conclusion is that Ethereum’s fundamental value is as a bulletin board. A place where anyone can write, anyone can read, and no one can modify or delete.

In the context of AI, this might be the most important statement Vitalik has made in the past two years.

We are entering an era where generative models become infinitely cheap. Text, images, videos, identities—all can be mass-produced by AI. When everything can be faked, what becomes scarce?

All these questions ultimately point to the same place: a public, persistent, irreversible data layer. And a record that no one can tamper with is exactly what Ethereum can provide.

Over the past two years, criticisms of Ethereum can be summarized as: What do you have that others can’t replace?

Looking now, Vitalik hasn’t directly answered this question.

However, over the past year, the Ethereum Foundation has done a few seemingly small things: assembled a 50-person privacy team, established a nearly 50-person privacy research group, released the Kohaku privacy framework, and appointed an AI lead; the 2026 roadmap prioritizes institutional privacy and faster transaction confirmation.

Looking back at his intense output over the past month, most of it revolves around privacy and efficiency issues for Ethereum in the context of AI.

I believe Vitalik is betting on one thing: the more powerful AI becomes, the more rigid the demand for privacy and verification infrastructure will be. Whether Ethereum can meet this demand is another matter, but he has clearly chosen his table.

ETH is still hovering around 2000. Most people still aren’t paying much attention to what he’s been saying lately.

But perhaps, in a few years, when looking back, the thing to care about will be this very period.