AI Tech / news
Open-Weight AI Model GLM-5.2 Nears Frontier Capabilities, Safety Gaps Persist
A SaferAI report finds Z.ai's open-weight GLM-5.2 is approaching frontier-level AI performance while lacking key safety mitigations, raising governance concerns.
The evaluation, conducted through Z.ai's public API, found that GLM-5.2 refused none of the offensive cyber or dual-use biology tasks it was given. By contrast, Anthropic's Claude Opus 4.7 refused so consistently that SaferAI could not complete CyberGym, a benchmark for cybersecurity capabilities, on it at all.
According to the report, GLM-5.2 is only a few months behind OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 on cyber and bio capabilities. This narrowing gap comes as policymakers debate how to govern increasingly powerful AI systems, including OpenAI's GPT-5.6 Sol and Anthropic's Mythos.
CyberGym is a benchmark used to evaluate cybersecurity capabilities; OpenAI applied it in the evaluation that preceded last month's Hugging Face breach. SaferAI's findings underscore a long-standing concern among critics that open-weight AI models could put highly capable technology into the hands of potential attackers, with no way to police its use once downloaded.
The report renews calls for stronger governance around open-weight models, as their capabilities grow faster than safety measures. SaferAI, a nonprofit focused on AI safety, has highlighted the growing divide between frontier capabilities and safety practices.