The Fable Recall Puts the Spotlight in the Wrong Place

The article argues that model jailbreaks and prompt‑injection attacks are inevitable because model refusals are statistical rather than enforceable; therefore defenders should stop treating the model itself as the primary security boundary. Instead, it recommends mapping a model's blast radius (tools, files, credentials it can reach) and placing deterministic or layered controls outside the model at leverage points so injected instructions cannot directly exfiltrate secrets or take harmful actions.