Vitalik Buterin Moves Away From Cloud AI Over Security Concerns

Vitalik Buterin is taking a very different approach to AI, and it’s already sparking conversations across both crypto and tech circles.

In a recent update shared here, Buterin revealed that he has completely stopped using cloud-based AI services. Instead, he now runs everything locally on his own machine.

It’s not just a personal preference move either. From what he explained, this decision is heavily tied to growing concerns around security, especially as AI agents become more powerful and autonomous.

Running A Full AI Stack Locally

What stands out immediately is how far he has gone with this setup.

Buterin is now running the Qwen3.5:35B model locally on an Nvidia 5090 laptop. That’s not a lightweight setup by any means, but it shows what’s becoming possible with current hardware.

According to the details shared, the system is capable of hitting around 90 tokens per second, which is quite fast for a local model of that size.

More importantly, everything runs directly on his device. No external servers, no cloud processing, and no data leaving the machine.

That level of control is a big part of the decision. For someone deeply involved in building decentralized systems, relying less on centralized AI infrastructure seems like a natural step.

Security Risks In AI Agents Becoming More Visible

The bigger issue Buterin pointed to is security.

He referenced research suggesting that about 15% of AI agent “skills” or integrations contain malicious instructions. That’s not a small number, especially when these agents are being used to automate tasks.

Even more concerning is the idea that something as simple as parsing a malicious webpage could compromise an AI assistant completely.

In other words, the risk isn’t always obvious. An agent might look like it’s doing something normal, while actually executing harmful instructions in the background.

As AI tools become more connected, to wallets, apps, and financial systems, those risks start to matter a lot more.

The Problem With Fully Autonomous AI

One of the key points Buterin seems to be making is about control.

There’s a growing trend of building AI agents that can act independently, send messages, execute transactions, interact with platforms, without needing constant human input.

On paper, that sounds efficient.

But according to his perspective, it can also be dangerous.

If an AI agent has the ability to move funds or make decisions without a human in the loop, it stops being just a tool. It becomes a potential liability.

And if that agent gets compromised, the damage can happen quickly, often before anyone notices.

That’s where his approach starts to differ from a lot of current AI development trends.

A More Conservative Approach To AI Interaction

To deal with these risks, Buterin is taking a more cautious route.

He has open-sourced a messaging daemon that requires human approval before any outbound message is sent to third parties. It’s a simple idea, but it changes how AI systems interact with the outside world.

On top of that, he is applying what’s described as a “2-of-2 confirmation” rule for sensitive actions. Basically, nothing critical happens unless there’s explicit human involvement.

It might sound slow or even unnecessary to some people, but that’s kind of the point.

Instead of optimizing for speed or full automation, the focus here is on safety and control.

And in an environment where AI systems are getting more powerful by the day, that trade-off might actually make sense.

Local AI Versus Cloud AI Debate Heats Up

This move also feeds into a bigger discussion around local versus cloud-based AI.

Running models locally gives users full control over their data and interactions. Nothing is sent out, nothing is processed externally. That reduces certain types of risks, especially around privacy and external manipulation.

But it’s not a perfect solution either.

Local AI doesn’t automatically mean safe AI. If the system itself is compromised, or if it processes malicious inputs, the risks are still there.

On the flip side, cloud AI offers convenience and scalability, but comes with its own concerns. When systems are connected to external servers and given broad permissions, users are essentially trusting those systems with a lot of control.

And as Buterin’s move suggests, that level of trust might not always be justified.

A Quiet Warning For The AI Agent Economy

Looking at the bigger picture, this feels like more than just a personal setup choice.

Vitalik Buterin is essentially pointing out a potential weak spot in the current direction of AI development, especially the push toward fully autonomous agents.

As more tools are built to act independently, the line between convenience and risk starts to blur.

His approach is almost the opposite of that trend. Less automation, more verification. Less external dependency, more local control.

It might not be the fastest or most exciting path, but it could end up being one of the more practical ones, especially for high-stakes use cases like finance or crypto.

For now, it serves as a reminder that as AI systems become more capable, the way they are designed, and controlled, matters just as much as what they can actually do.

Disclosure: This is not trading or investment advice. Always do your research before buying any cryptocurrency or investing in any services.

Follow us on Twitter @themerklehash to stay updated with the latest Crypto, NFT, AI, Cybersecurity, and Metaverse news!