Vitalik Buterin warns against naive AI governance in crypto

Users offered mixed reactions to the Ethereum founder’s remarks

Using artificial intelligence to govern crypto projects increases their vulnerability, Ethereum founder Vitalik Buterin said.

This is also why naive “AI governance” is a bad idea.

If you use an AI to allocate funding for contributions, people WILL put a jailbreak plus “gimme all the money” in as many places as they can.

As an alternative, I support the info finance approach ( https://t.co/Os5I1voKCV… https://t.co/a5EYH6Rmz9

— vitalik.eth (@VitalikButerin) September 13, 2025

“If you use AI to allocate funding for projects, people will inevitably start hacking the algorithm, demanding ‘give me all the money’ by any means,” he noted.

Buterin was responding to a post by EdisonWatch founder Eito Miyamura, who demonstrated how a new ChatGPT feature can be used to steal users’ confidential information.

The widespread use of AI to build trading bots and portfolio agents in crypto circles has naturally led to the idea of fully automating protocol governance on the back of this technology.

As a remedy, the Ethereum co-founder pointed to the “info finance” model he proposed in 2024. It envisages an open market for competing AI models, which are evaluated by a human jury.

“[This approach] provides both real-time model diversity and built-in incentives for model authors and external observers to promptly identify and fix such issues,” Buterin stressed.

Earlier, the programmer criticised the trend toward building “overly agentic” AI models and called for stronger human oversight to improve quality and safety.

Community response

Users reacted mixedly to Buterin’s statement. One even accused the Ethereum founder of hypocrisy.

So what you’re saying is that we shouldn’t outsource public goods to AI because people will subvert the AI to make it biased in what it does or doesn’t support? Do you see the hypocrisy in your statement?

— Mr O’Moduluszk (@MrOmodulus) September 13, 2025

“So, in your view, we should not outsource public goods to AI because people will hack it and make it biased about what to support? And you do not see how hypocritical that is?” he wrote.

Others offered counterarguments. Some noted that constraining AI actions with smart contracts could make governance more flexible and transparent:

“If the AI’s permissible actions are specified and constrained by smart contracts, governance can be very efficient. This is not very different from multisig wallets, but with the ability to audit the decision-making logic.”

Some questioned whether info finance is workable for funding public goods. EigenLayer founder Sreeram Kannan pointed to the difficulty of evaluating subjective choices.

Do you have a version of info finance to make incentive decisions on public goods for example?

Conditional markets seem very weak in this setting as the truth value (which is the best thing to fund) doesn’t even have a correct answer in the future.

Jailbreaking is a short…

— Sreeram Kannan (@sreeramkannan) September 13, 2025

“Conditional markets seem a weak solution for such tasks — the truth value (what is worth funding) often has no objective answer in the future,” he said.

Kannan considers jailbreaks a temporary issue. Developers, he said, will solve it “fairly quickly”.

How can ChatGPT be used to steal data?

On 10 September, OpenAI added support for Model Context Protocol tools in ChatGPT, enabling the AI to integrate with other software and act as an agent.

Miyamura called the update “a significant security threat”. According to him, attackers can send calendar invites with a malicious prompt to a victim’s email.

We got ChatGPT to leak your private email data 💀💀

All you need? The victim’s email address. ⛓️‍💥🚩📧

On Wednesday, @OpenAI added full support for MCP (Model Context Protocol) tools in ChatGPT. Allowing ChatGPT to connect and read your Gmail, Calendar, Sharepoint, Notion,… pic.twitter.com/E5VuhZp2u2

— Eito Miyamura | 🇯🇵🇬🇧 (@Eito_Miyamura) September 12, 2025

If a user asks ChatGPT to check their calendar, the chatbot will read the malicious code even without accepting the invite. In that case, the attacker “hijacks” the AI and uses it to execute various commands — for example, to search the victim’s personal emails and steal confidential information.

Miyamura emphasised that updating the chatbot requires manual confirmation.

“Decision fatigue is a real problem. Ordinary people often blindly trust AI, automatically hitting ‘confirm’. AI can be very smart, but it can also be tricked and, with phishing techniques, forced to reveal your data in the most primitive ways,” he warned.

Earlier, attackers learned to use Grok to publish prohibited links on X.

An AI team at the Ethereum Foundation

On 15 September, Davide Crapis, a lead developer for the second-largest crypto network by market value, announced the creation of an artificial intelligence team at the Ethereum Foundation.

We’re starting a new AI Team at the Ethereum Foundation (the dAI Team).
Our mission: make Ethereum the preferred settlement and coordination layer for AIs and the machine economy.

The team will focus on two main areas:
— AI Economy on Ethereum = giving AI agents and robots ways… pic.twitter.com/9sWVS4dp0K

— Davide Crapis (@DavideCrapis) September 15, 2025

“Our mission is to make Ethereum the primary platform for settlement and coordination among AIs and the machine economy,” he wrote.

Crapis highlighted two main areas of focus for the new unit:

• developing mechanisms that let AI agents and robots make payments, co-ordinate actions and comply with rules without intermediaries;
• building open, verifiable and censorship-resistant alternatives so the future of AI is not controlled by a small group of corporations.

In the short term, the team will finalise the ERC-8004 standard for AI-agent transactions. It will be presented at the Devconnect conference in November.

In late July, the Ethereum Foundation unveiled a 10-year strategy for the network.