Stopping the pile-on before it makes things worse
A circuit breaker monitors calls to a dependency and, after enough consecutive failures, stops sending it new requests for a while, rather than letting every caller keep retrying against a system that's already struggling. Continued retries against a failing dependency don't just fail themselves — they can consume the exact capacity the dependency would need to recover, turning a temporary problem into a sustained outage.
Applied to a policy service specifically, tripping the breaker open removes load from a struggling dependency at precisely the moment removing load matters most, giving it a real chance to stabilize instead of being overwhelmed by a growing backlog of retried authorization checks.
The breaker doesn't answer what happens during the outage
Once the breaker is open and the real policy service isn't being called at all, something still has to happen for every request that would have gone to it — and that's exactly the fail-closed-versus-fail-open question, asked again in a new form. A circuit breaker changes when the fallback kicks in; it has nothing to say about what the fallback should actually do, which has to be decided deliberately and explicitly, the same way it would for any other authorization dependency failure.