Anthropic published its Responsible Scaling Policy in September 2023, the first major lab to commit publicly to capability evaluations and safety thresholds before training larger models.