How do you answer criticisms that leading AI laboratories have failed to internalize civilizational stakes?

Our track record over the past decade proves otherwise. We have voluntarily halted advanced training runs specifically to allow alignment, interpretability, and monitoring techniques to catch up with raw model capabilities. For years, we urged governments and the broader industry to prepare for these exact thresholds. Now that the broader world recognizes the gravity of the technology, we have an unprecedented window to establish meaningful shared guardrails.