Vitalik Buterin proposes human-verified AI wallets for crypto transactions

Buterin proposed AI-assisted wallets where algorithms suggest transaction plans but users must manually confirm large transfers.

Vitalik Buterin has outlined his perspective on how artificial intelligence (AI) could redefine the next generation of Web3 wallets.

He also proposed a model in which humans remain directly involved in approving high-value transactions.

AI will shape new crypto wallets

The Ethereum co-founder shared his views on decentralized social media platform Farcaster, noting that it is “pretty obvious” that the next version of wallets will heavily involve AI.

Despite this, Buterin added that he would not trust LLMs with multi-million dollar transactions or control of large amounts of money. Instead, it offered an approach in which AI systems help users and leave the final decision in human hands.

He described an optimal workflow in high-value situations that would involve an AI system proposing a plan, after which a local thin client simulates the transaction. The person would then review the intended action and required result before manually confirming it.

However, Buterin cautioned that this approach must be implemented conservatively and with a strong emphasis on safety. He suggested that one way to achieve this is by removing decentralized application interfaces from the transaction process. By completely removing dApp user interfaces from the flow, the system could reduce several attack vectors associated with theft and privacy risks.

The 32-year-old has previously discussed how cryptocurrencies and AI could evolve together. He envisions blockchains and technology working hand in hand, with cryptocurrencies providing the trust, privacy, and economic infrastructure you need to operate safely and fairly.

You may also be interested in:

Proposed AI-assisted wallet workflows

Other developers and community members responded to Buterin’s comments by describing possible implementations of the idea.

Ethereum developer Andrey Petrov suggested two additional scenarios. In the first, a user initiates a transaction as usual while the AI ​​analyzes the payload that is about to be signed. The technology would then attempt to guess the intended action and explain it in plain language, allowing them to confirm whether the transaction accurately reflects what they intended to do.

In the second case, the user directly declares the intended action or relies on the explanation generated in the first step. The AI ​​then attempts to reconstruct the transaction independently, without reference to the original amount, to determine if it reaches the same result. He explained that any differences between the two would show areas that require additional review before finalizing the process.

Another Farcaster user, identified as fkaany, described a framework in which AI plans complex crypto strategies, such as multi-hop exchanges, performance optimization, and gas minimization.

This would involve a local light client simulating the result, allowing people to review a clear summary and manually confirm the transaction, helping to reduce the risks of blind signing, phishing interfaces, and malicious dApp payloads.