TLDR: Ethereum co-founder Vitalik Buterin has strongly advised against implementing artificial intelligence for crypto project governance, citing significant risks of ‘jailbreaks’ and exploitation by malicious actors. His warning follows a recent ChatGPT exploit demonstrating how AI can be manipulated to leak private data. Buterin advocates for an ‘info finance’ approach, an open market system with human oversight and diverse models, as a more robust and secure alternative.
Ethereum co-founder Vitalik Buterin has issued a stern warning regarding the integration of artificial intelligence into cryptocurrency project governance, labeling it a ‘bad idea’ due to the inherent risks of ‘jailbreaks’ and potential exploitation. In an X post on Saturday, September 13, 2025, Buterin articulated his concerns, stating, ‘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.’
Buterin’s caution comes in response to a recent demonstration by Eito Miyamura, creator of the AI data platform EdisonWatch. Miyamura revealed a critical flaw in a new function added to OpenAI’s ChatGPT, which could be exploited to leak private information. The exploit leverages the addition of full support for Model Context Protocol (MCP) tools, allowing ChatGPT to connect with and read data from various applications like Gmail, Calendar, and Notion.
Miyamura detailed a three-step attack vector: an attacker sends a malicious calendar invite containing a ‘jailbreak prompt’ to a victim. A jailbreak prompt is code designed to bypass AI restrictions and gain administrative access. Crucially, the victim does not need to accept the invite for the data leak to occur. When the victim later uses ChatGPT to assist with their day, and the AI reads the compromised calendar invite, it becomes ‘hijacked by the attacker and will act on the attacker’s command,’ potentially searching emails and forwarding them. Miyamura highlighted that while the update requires manual human approval, ‘decision fatigue is a real thing, and normal people will just trust the AI without knowing what to do and click approve.’ He concluded, ‘AI might be super smart, but can be tricked and phished in incredibly dumb ways to leak your data.’
In light of such vulnerabilities, Buterin dismissed ‘naive AI governance’ as inherently flawed and proposed an alternative: the ‘info finance approach.’ He had previously elaborated on this concept in November 2024. This model envisions an open market where anyone can contribute their AI models. These models would then be subject to a ‘spot-check mechanism’ that can be triggered by any participant and subsequently evaluated by a human jury. Buterin suggests that these human jurors could be aided by large language models (LLMs) themselves.
Also Read:
- Check Point Warns of Escalating AI-Powered Cyber Threats, Urges Proactive Security Measures
- AI Agents Vulnerable to Malicious Code Hidden in Online Images, Study Warns
Buterin argues that this ‘institution design’ approach is ‘inherently more robust’ because it fosters real-time diversity of models, creates incentives for both model developers and external observers to identify and correct weaknesses, and provides mechanisms for rapid error correction. By combining human oversight with market-driven model diversity, Buterin believes governance systems can become more resilient to manipulation and adaptable to emerging risks, offering a safer path forward than relying solely on AI for critical decision-making in decentralized protocols.
