How Far Behind the Frontier are Leading Open Weight Models on Cyber?
Summary
AISI's frontier AI trends analysis shows open-weight models GLM-5.2 and DeepSeek V4-Pro are closing the gap to frontier closed models, lagging by about 4–7 months. The article discusses benefits and risks of open-weight models, especially around safeguards, cost, and defender readiness, and highlights getting ahead of potential misuse.