Ethereum co-founder Vitalik Buterin has outlined how he thinks next-generation AI crypto wallets should be built, and his line is clear: AI should not be trusted to move large sums on its own. In his view, an AI wallet should behave like an assistant that understands intent and prepares a transaction plan, while the user keeps final control over any transfer of funds.
The idea is aimed less at blockchain consensus and more at the risks people face during everyday interaction. A large share of crypto activity still runs through dApp interfaces, where fake websites, copied pages, and misleading signing prompts remain common. Buterin’s proposed model shifts much of that interaction back into the wallet itself. Instead of visiting a website and clicking through uncertain prompts, a user could tell the wallet what they want to do and let the AI handle the chain interaction path.
A three-step flow keeps the human in control
Buterin described a simple safety flow with three parts. First, the AI reviews the user’s goal and creates a transaction plan. Second, a local light client tests that plan before anything is signed, showing what the outcome would look like, including changes to balances. Third, the result of that simulation appears on the user’s screen, and the user manually confirms only if the outcome looks correct.
The point is not automation for its own sake. It is a strict limit on authority. AI can interpret instructions, organize steps, and present a suggested action path, but it should not skip the approval process or execute transfers without permission. That keeps the difficult technical work on the machine side while the final decision stays with the asset holder.
Local simulation is treated as the key safeguard
The local light client is central to the design. Rather than relying on a large company or remote intermediary to validate what will happen, the wallet would run a test on the user’s own phone or computer. What the user sees is not a vague warning but a concrete preview of the transaction’s likely effect: how balances change and what the action actually does. In short, simulate first, sign later.
That approach also fits crypto’s decentralization goals. Because the crucial checking step happens on the local device, the user does not have to hand over trust to a third party simply to understand a transaction. For people using DeFi, sending tokens, or managing more complex onchain actions, the model is designed to replace blind signing with a review process tied to visible outcomes.
The wallet could reduce exposure to risky dApp websites
Buterin also pointed to dApp websites as a major attack surface. Many losses in crypto do not begin with protocol code failure but with websites that look legitimate and persuade users to sign something they do not fully understand. Under his proposal, next-gen AI wallets could reduce or even bypass that website layer by turning “go to a page and click buttons” into “state the intended action inside the wallet.”
That would make the experience closer to a guided conversation than a chain of approvals scattered across browser tabs. The complexity does not disappear, but AI could absorb much of it before presenting the user with a final review. Ease of use improves. The important caveat is that control does not move with it.
Better usability, but no full delegation to AI
The source article notes that DeFi interactions and token transfers still feel difficult for many users. AI wallets are presented as a way to make those actions easier to understand and complete. But Buterin does not frame AI as a fully trusted custodian. He keeps returning to manual review, especially where large amounts of money are involved.
The article also says that by the end of 2026, more wallet developers may begin adding these features. The stated goal is a wallet that feels as easy to use as a banking app while keeping a high security standard. It closes with a standard risk reminder: crypto assets are volatile, and seed phrases should never be shared with any software or person.