Related articlesactivation probesLightweight L0 classifiers that scan model internal activations for concerning patterns.OpenAI Expands Layered Monitoring to Detect Frontier Model Misalignment After Hugging Face Incident