Saturday, 24 October, 2026
Safety First: Moonshot Dulls Kimi's Hacking Edge Amid Rising AI Threat Fears
By TechShots Studio
Moonshot AI deliberately filtered cyber-related training data for its new Kimi K3.1 Pro model to curb its offensive hacking abilities. While an unrestricted version scored 90.5% on internal vulnerability benchmarks, the safety-filtered variant dropped to 81.0%. This rare public move highlights growing industry pressure to preemptively restrict powerful AI models from being weaponized for automated cyberattacks.
Read full story at REUTERS