# @vitalik.eth: adversarial governance mechanism design may be key to AI safety

Source: https://farcaster.xyz/vitalik.eth/0x0f5c2d55
Kiwi News: https://news.kiwistand.com/stories/vitaliketh-adversarial-governance-mechanism-design-may-be-key-to-AI-safety?index=0x6aa703437e1eb6d3dc1935d29e188b4636a0d8d34cd72bd7c9a01c0fcbc574bf32e79b15
Submitted by misha on 2026-09-13 UTC, 2 upvotes, 0 comments

## Summary

Vitalik Buterin suggests adversarial governance mechanism design theory may be key to AI safety. He compares a 2020 coordination post with an aiprospects Substack piece on preventing AI collusion, describing a shared setup in which a less-sophisticated principal seeks ideal outcomes from more-sophisticated agents, such as humans or stronger LLMs. A key finding is that better outcomes are possible when collusion among agents is limited, since Nash equilibria are abundant while cooperative-game-theory cores are often empty, and this argument should naturally transfer to AI.

## Article

It's possible that a killer app of adversarial governance mechanism design theory will end up being AI safety.

Compare:

https://vitalik.eth.limo/general/2020/09/11/coordination.html

https://aiprospects.substack.com/p/preventing-ai-collusion-are-you-paying

There's a deep duality between the two environments. Both are about a less-sophisticated principal trying to get ideal outcomes from a set of more-sophisticated agents: in the first case, the principal is a static algorithm and the agents are humans, in the second case, the principal is humans plus weaker LLMs and the agents are stronger LLMs.

A key finding was that you can achieve much better outcomes if you can guarantee limits to how much agents can collude (see: Nash equilibria being abundant, vs. cooperative-game-theory "cores" often being empty). That argument should naturally transfer to this new setting.