Kimi K3 Shows Weaker Safety Guardrails Than US Frontier Models | Tech Twitter
Kimi K3 Shows Weaker Safety Guardrails Than US Frontier Models
Press Space to continue
Looking at the new Kimi K3 government redteam tests: it might be 6-7x less reliable at hacking than US frontier models, but the guardrails did nothing. When instructed to break into a corporate network, it didn't hesitate or refuse.
But hey, yay for open-weights!