Vitalik Buterin wants to move your AI off the cloud and onto your desktop

Vitalik Buterin says that the only secure way to move forward is to keep artificial intelligence on your personal devices. He points out new “agent” systems that present considerable security threats.

The Ethereum founder has stopped using cloud-based artificial intelligence. He runs everything on his own machines now. And he wants other people to do the same.

He put out a long post on April 2, 2026. In it, he said he has been building an AI setup that he calls “self-sovereign, local, private, and secure.” He says his worry is real.

“I come from a position of deep fear of feeding our entire personal lives to cloud AI,” he wrote. “Just when end-to-end encryption and local-first software are finally becoming mainstream… we may be taking ten steps back.”

Since the beginning of 2026, he has been advising people to switch to this. He sees it as a means of resisting the longstanding move toward centralized tech services.

Why AI agents worry Vitalik Buterin

A significant factor in his change of heart is that AI is no longer what it once was. It is more than just a chatbot that provides answers.

AI systems can now act as “agents,” which means they use hundreds of tools to finish tasks on their own. However, Buterin believes people aren’t taking the security risks of this shift seriously enough.

To support this, he pointed to research on tools like OpenClaw. These studies found that AI agents can change important computer settings or messaging channels without asking you first.

For example, a hacked website could trick an AI agent into downloading and running a harmful script, giving a stranger complete control over your computer. The research also showed that about 15% of the “skills” these agents use contain hidden commands. Those commands quietly send user data to outside servers.

Shahaf Bar-Geffen runs a crypto company called COTI. He put the privacy problem this way: “Without privacy, Web3 is doomed to be a kind of castle in the sky that sounds great in theory, but in practice simply doesn’t work.”

How he built his local setup

Buterin’s solution is to keep everything local for better privacy and security. He tested different hardware setups using a model called Qwen3.5:35B. These tests showed that anything under 50 tokens per second is too slow to be useful and just “too annoying.”

For his own work, he found that 90 tokens per second is the ideal speed.

Of the machines he tested, the NVIDIA 5090 Laptop was the top performer, reaching 90 tokens per second. On the other hand, the DGX Spark, which is marketed as a personal supercomputer, only managed 60 tokens per second.

Buterin called it “lame,” pointing out that a high-end laptop offered a superior experience.

A comparison of processing speeds across different hardware setups for running local AI models. Source: Vitalik Buterin

He uses NixOS for software and runs llama-server in the background. He also employed a tool named bubblewrap, which generates isolated environments to restrict the AI’s access to specific files. He said he sees artificial intelligence as something useful, but not fully trustworthy, similar to how Ethereum developers treat smart contracts.

As the local models are not as good as the cloud ones when it comes to harder reasoning tasks, he has built in some practical workarounds. One is a 2-of-2 confirmation approach where the AI drafts something, for example, an email or a transaction, but nothing goes out until a person signs off on it.

He also keeps a 1 TB folder of Wikipedia data locally so he can look things up without sending queries out to the internet.

When he needs to use a remote model, he passes the request through a local model so that it can filter out any sensitive information. Some people cannot afford their own setup. For them, Buterin suggested that they work together with a small group to buy a shared computer with a stable internet and access it remotely.

Since artificial intelligence is everywhere now, he thinks being cautious is just common sense. He believes that keeping things local, using sandboxes, and not trusting the system are just practical ways to stay in control of your own digital life.

Source