=

The passage around this formula

It helps to split the risk in an enterprise AI deployment into three layers, governed by different mechanisms, audited by different parties, and owned by different organizations: $R_{\mathrm{enterprise}} = R_{\mathrm{model}} + R_{\mathrm{platform}} + R_{\mathrm{application}}$. $R_{\mathrm{model}}$ is the risk that the underlying weights behave dangerously or unpredictably at the frontier of capability. This is the layer Anthropic’s Responsible Scaling Policy is built to govern: it commits Anthropic to defined AI Safety Level thresholds, to Frontier Safety Roadmaps naming specific safety goals, and to Risk Reports quantifying risk across deployed models, with a designated Responsible Scaling Officer accountable for the policy’s implementation [ 7 ] .…

Read this part in the article →