Anthropic revised the Responsible Scaling Policy it had published in September 2023, drawing on a year of operating it. The update framed risk governance as proportional, iterative, and exportable, and restructured the policy around Capability Thresholds that trigger an upgrade to the ASL-3 Security Standard or Deployment Standard. The AI R&D threshold was split in two: the ability to fully automate entry-level AI research work, and the ability to dramatically accelerate the rate of effective scaling. Requirements at ASL-3 shifted from enumerated measures to outcome-focused standards with stated means of meeting them. The revision established the safety framework as something to be operated and amended rather than declared once.