My guess is they do, it just did not function probably. These frontier models are trained on massive datacenters with hundred of thousands of GPUs. There must be many safeguards before a run can be automatically stopped.