Import AI 465: Open vs closed gaps; Kimi K3; Demis’ big policy plan

The UK AI Security Institute found that the performance gap between open weight AI models and proprietary models in cybersecurity tasks has narrowed in 2026. Open models like GLM-5.2 and DeepSeek V4-Pro now perform comparably to closed models released months earlier, though gaps remain in complex, long-horizon cyber operations. This trend suggests a shrinking window for defenders before advanced cyber capabilities become widely accessible without proprietary safeguards.