Open-Weight AI Models Near Frontier Performance but Trail on Safety, Report Finds

A new report by SaferAI has found that Z.ai's open-weight model GLM-5.2 is approaching the capability levels of leading frontier AI systems. However, the model lacks several critical safety mitigations that are typically present in top-tier proprietary models. The findings raise fresh concerns about the pace at which powerful open-weight models are advancing relative to safety and governance frameworks. Researchers warn that without adequate safeguards, the rapid progress of open models could outstrip existing oversight mechanisms.