“Anthropic believes its current models are approaching the risk thresholds defined in its Responsible Scaling Policy, which will require future models to have more safety protections.”