Vitalik Says Use-Based AI Limits Are Less Harmful Than Identity-Based Treatment

Vitalik said restrictions on AI capabilities based on how they are used would be less harmful than identity-based treatment. He said society needs time to prepare for powerful cyberattacks.

He said it was acceptable for Kimi 3.1’s training data to exclude advanced offensive cyber capabilities. He said preparation should include strengthening systems, reviewing older code and applying formal verification to new code.

Vitalik said a poor response to the first major AI risks could lead to highly restricted models, widespread know-your-customer checks and surveillance. In that scenario, users could receive different levels of model access based on their credentials and nationality.