Vitalik: Restricting AI capabilities by use case is better than restricting by user identity

Odaily reports: Vitalik posted on X platform that he believes it is acceptable for Kimi 3.1 to intentionally exclude high-end offensive cyber capabilities from its training data. Compared to treating users differently based on identity, treating them differently based on use case is far less problematic; society needs time to prepare, harden systems, review old code, and formally verify new code before a full-scale wave of powerful cyberattacks arrives. Vitalik also stated that if the first major wave of AI risk is not properly addressed, society could move toward a scenario of highly restricted models, widespread KYC and surveillance, where users might be divided into different tiers based on credentials and nationality, receiving different levels of model access.