OPEN AI MODELS GAIN POWER, SAFETY LAGS BEHIND

Open-weight AI models—those with publicly available code and weights—are narrowing the performance gap with proprietary frontier systems. Z.ai's GLM-5.2 demonstrates capability parity with leading closed models, but the SaferAI report identifies significant safety deficits. The analysis highlights missing safeguards in areas including adversarial robustness, bias mitigation, and alignment with human values. As open models grow more capable, the disparity between raw performance and safety measures creates potential risks. The issue intensifies existing tensions in AI governance. Open-source advocates argue transparency drives innovation and security; critics contend that releasing powerful systems without robust safety measures accelerates risks faster than the regulatory landscape can address. The gap underscores a central challenge: balancing the benefits of open AI development with the need for responsible deployment of increasingly capable systems. Industry observers warn that without coordinated safety standards, the advantage may tip toward speed over security.