Vitalik backs Kimi 3.1 excluding top offensive cyber capabilities from training data
JiaweiLiu_ · x · 2026-10-26
Vitalik Buterin endorsed Kimi 3.1's decision to deliberately leave out top-end offensive cyber capabilities from its training data.
His core argument: discriminating against use cases is much less bad than discriminating against users, and the world needs all the time it can get to prepare — hardening old code and (properly) formally verifying new code — before a full wave of powerful cyber attacks arrives.
He also warns that mishandling this first wave of AI risk could lead to a dystopia of heavily locked-down models, KYC gating everywhere, and intrusive surveillance where model access depends on credentials and nationality.
More from Models
- Ternary 1.58-bit Nemotron 30B-A3B GGUF Runs 368K Context on 12GB GPU — pmttyji · 2026-10-26
- AI researchers lament models refusing vulnerability analysis of old PHP code despite 'use AI to defend' messaging — voooooogel · 2026-10-26
- Developer confirms models are systematically refusing Apache-related vulnerability requests — jd_pressman · 2026-10-26
- Dev shares $5,107 monthly AI bill: 10B tokens, DeepSeek cost $0.70 — cneuralnetwork · 2026-10-26
- Skipping eval checks means you're likely fitting benchmark biases and noise — antoine_chaffin · 2026-10-26
- DeepSeek V4.1-Flash hits 49 tok/s decode on dual Radeon R9700 in local test — Longjumping-Elk-7756 · 2026-10-26