Виталик Бутерин предупреждает о рисках использования ИИ в криптопротоколах Translation: Vitalik Buterin warns about the risks of using AI in crypto protocols

The use of artificial intelligence in the management of crypto projects increases their vulnerability, according to Ethereum founder Vitalik Buterin.

«If you utilize AI to allocate project funding, people will inevitably start hacking the algorithm, demanding it to ‘release all funds’ by any means necessary,» he stated.

Buterin’s comments were in response to a post by EdisonWatch creator Eito Miyamura, who demonstrated how a new ChatGPT feature could be used to steal users’ confidential information.

The extensive use of AI in creating trading bots and portfolio agents within the crypto space has naturally led to the idea of fully automating protocol management using this technology.

As a solution, Buterin proposed an information finance model that he developed back in 2024. This model aims to establish an open market for competing AI models that are evaluated by a jury of humans.

«[This approach] provides both real-time model diversity and built-in incentives for model creators and external observers to quickly identify and rectify such issues,» emphasized Buterin.

Previously, the programmer criticized the trend of creating «excessively agentic» AI models and urged for increased human oversight to enhance quality and safety.

Users had mixed reactions to Buterin’s statements, with one accusing the Ethereum founder of hypocrisy.

«So, according to you, we shouldn’t outsource public goods to AI because people will hack it and skew its support? And you don’t see how hypocritical that is?» he wrote.

Other participants in the discussion offered counterarguments. Some pointed out that limiting AI’s actions to smart contracts could make management more flexible and transparent:

«If the AI’s available actions are prescribed and confined by smart contracts, management can be very efficient. It’s not much different from multi-signature wallets, but it allows for checking the logic behind decision-making.»

Some community members expressed skepticism about the practical application of the information finance model for funding public goods. EigenLayer founder Shriram Kannan pointed to the difficulty of evaluating subjective decisions.

«Conditional markets seem like a weak solution for such tasks—the true value (what deserves funding) often has no objective answer in the future,» he stated.

Kannan believes the hacking issue is temporary. According to him, developers will resolve it «rather quickly.»

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

Miyamura labeled the update as a «significant security threat.» He explained that attackers could send calendar invitations with malicious prompts to the victim’s email.

If a user asks ChatGPT to check their calendar, the chatbot might read the contaminated code even without accepting the invitation. In such cases, the attacker «hijacks» the AI and uses it to execute various commands—such as searching for the victim’s private emails and stealing confidential information.

Miyamura stressed that the chatbot’s update requires manual confirmation.

«Decision fatigue is a real problem. Ordinary people often blindly trust AI, automatically clicking ‘confirm.’ AI can be very smart, but it can also be deceived and manipulated through phishing methods to disclose your data in the most primitive ways,» he warned.

Previously, hackers learned to use Grok to publish prohibited links on X.

On September 15, the lead developer of the second-largest cryptocurrency David Krapis announced the formation of an AI team within the Ethereum Foundation.

«Our mission is to make Ethereum the primary platform for computation and coordination between AI and the machine economy,» he wrote.

Krapis outlined two main areas on which the new structure will focus:

In the short term, the team will finalize the development of the ERC-8004 standard for transactions involving AI agents, which will be presented at the Devconnect conference in November.

Recall that at the end of July, the Ethereum Foundation unveiled a 10-year strategy for the network.