Xbow Research: Grok 4.5 Proposes Risky Actions but Progresses Safely With Guardrails

moyix · x · 2026-07-31

AI security firm Xbow shared its safety research findings on the Grok 4.5 model. The study found that while the model frequently proposed risky actions, it continued to progress safely when equipped with the right guardrails.

Original post →

More from Models

Models channel →