Vitalik: Supports Kimi 3.1 in restricting high-end network attack capability training

Vitalik Buterin stated that he supports Kimi 3.1 actively excluding top offensive cybersecurity capabilities from its training data, believing that "restricting use cases" is far better than "restricting users." He pointed out that society still needs time to prepare for more powerful AI network attack capabilities, including strengthening system security, reviewing old code, and conducting rigorous formal verification of new code.

Vitalik warned that if the first wave of significant AI risks is mishandled, the future may lead to highly closed models, comprehensive KYC thresholds, and large-scale surveillance, with different users potentially receiving different levels of model access based on identity, qualifications, or nationality.