They are intentionally failing to implement proper security measures to be able to claim that further AI development would be dangerous. What it would actually be is expensive, and AI firms want to stop spending on R&D without worrying about competitors (particularly open weight competitors) leapfrogging them.