Vitalik Warns AI Security Could Concentrate Power

Vitalik Buterin the founder of Ethereum, warned that efforts to manage the risks of artificial intelligence (AI) could concentrate power in a small group, calling instead for defensive tools that people can use and verify for themselves.

In his Devcon keynote, he described a model of security in which one organisation holds the most powerful tools while others are denied access. He argued that this approach could encourage a race as those left outside seek the means to protect themselves.

The session, titled The Open Secure and Verifiable Half to a Brighter Future, took place on 3rd November at Devcon 8 in Mumbai. Bored Node’s earlier Devcon report covers the gathering’s focus on Ethereum, privacy and open technology.

Security without concentrating control

Buterin framed the challenge around three outcomes he wanted to avoid: stopping technological progress out of fear, concentrating control in a powerful organisation, or allowing dangerous technology to cause catastrophic harm.

His alternative was to accelerate defensive technology and make it openly available. He argued for preventing harm across digital systems, information and health, with people able to inspect, improve and share the tools they depend on.

That position also left room for restraint. Buterin said the most advanced AI models should not progress so quickly that there is insufficient time to develop protective measures.

He paired openness with verifiability: users should be able to check whether a system has the properties its makers claim. He pointed to formal verification, which uses logic to establish specified properties of hardware or software designs, and reducing unnecessary dependence on external servers.

Keeping private information out of AI requests

Buterin described AI as increasingly becoming a second brain, raising concerns about private thoughts passing through a company’s servers, where they could be exposed through hacking or used for other purposes.

He proposed using AI models on a user’s own device where possible. When a more capable remote model is needed, he suggested limiting its authority and using a local model to remove private information before sending a request. He also called for checks on responses coming back from remote systems.

These were areas where he said more work was needed to combine powerful AI with the privacy and security of local systems.

Privacy ran through his wider argument. He urged developers to avoid collecting data where possible and pointed to cryptographic proofs that establish a specific fact without requiring someone to disclose all the underlying personal information.

Beyond AI models

Buterin extended the same demand for verifiable security to operating systems and hardware. His examples included televisions that users cannot confidently assess for surveillance and hardware wallets whose components need to be trustworthy. Our wallet guide explains the distinction between a wallet interface and the authority to spend assets.

For messaging, he called for open software, verifiable security and protection for metadata, including who communicates with whom and when. For social media, he argued that users need meaningful control over what they see and practical ways to leave or create communities.

His proposed changes went beyond publishing software code. They included making different social platforms work together, giving users more control over moderation, and making it feasible to establish a community without building an entirely new set of infrastructure.