UK AI Security Institute Finds Open Models Lag Frontier Models by 4-7 Months in Cyber Capabilities
The UK AI Security Institute's first public analysis reveals open models like GLM-5.2 and DeepSeek V4-Pro lag frontier models by 4-7 months on long-horizon cyber range tasks. This gap has narrowed from 6-10 months through most of 2025, indicating advanced capabilities are reaching less safeguarded open models faster.
So What
The narrowing gap between open and frontier model cyber capabilities suggests open-source models are becoming more competitive for security-sensitive applications.