Vitalik says limiting AI by use case is preferable to restricting access by identity

Ethereum co-founder Vitalik said in a post on X that he considers it acceptable for Kimi 3.1 to intentionally exclude advanced offensive cyber capabilities from its training data. In his view, restrictions based on intended use carry far fewer downsides than systems that treat users differently based on who they are. He argued that society still needs time before powerful cyberattacks become widespread, including time to harden systems, review old code, and apply formal verification to new code. Vitalik also warned that if the first major wave of AI risk is handled poorly, the result could be a much more restrictive environment for model access. That could include heavily constrained models, broad KYC requirements, and expanded surveillance, with users sorted into different tiers based on credentials and nationality and given different levels of access.