Open AI Models Catching Up to Frontier in Capability, But Lack Safety Guardrails
RebeccaBellan · x · 2026-08-05
As policymakers debate AI regulation, a new report by nonprofit SaferAI reveals that the Chinese open-weight model GLM-5.2 is narrowing the capability gap with frontier closed models from OpenAI and Anthropic in cyber and biosecurity, yet the safety practice divide is widening.
Evaluations show GLM-5.2 refused none of the offensive cyber or dual-use biology tasks assigned, whereas Anthropic's Claude Opus 4.7 refused so consistently that evaluators couldn't even complete the cybersecurity benchmark. Furthermore, the White House's voluntary AI eval framework currently excludes open-weight models, leaving a critical safety gap as open capabilities accelerate.
Related event: Report: GLM-5.2 Nears Frontier Capabilities but Lacks Safety Guardrails(3 posts)→
More from Models
- Shieldstral: 3B-Parameter Open Multimodal Safety Classifier — petrusenko_max · 2026-08-05
- Xiaomi Launches Robotics-1: A New Foundation Model for Robots — AdinaYakup · 2026-08-05
- Fatal Flaws in Top AI Coding Models Open Door for a Strong Third Player — robleclerc · 2026-08-05
- Healthcare Agent Test: Most Expensive Models Make the Most Mistakes — Ubunta · 2026-08-05
- Andrew Ng Says Chinese Open-Weight Models Are Safer Than Closed-Source — pstAsiatech · 2026-08-05
- Redefining Frontier Models: OpenAI Leads the Pareto Frontier — downingARK · 2026-08-05