Score any agent deployment with four questions before you ship it
Claude by Anthropic·high signal
Anthropic Deputy CISO Jason Clinton published a four-question triage framework on July 17, 2026: what untrusted content does it ingest, what actions can it take and on whose behalf, what is the blast radius if it is misaligned, and what observability do I have to distinguish agent actions from user actions. The framing is explicitly that zero risk is not the goal — making risk legible and bounded is. It is a 10-minute exercise that converts hand-wavy 'is this agent safe' arguments into a scored, comparable answer per use case.