Minimum-necessary permission model with time-bound grants

Control objective

A minimum-necessary permission model ensures each agent holds only the access it genuinely needs, only for as long as it needs it - the principle of least privilege applied to autonomous agents. Over-permissioned or long-lived standing access is a prime target: if an agent is compromised, broad or stale grants let an attacker reach far more bank systems and data than the task ever required. To implement, scope every agent's permissions tightly to its task, issue access as short-lived, time-bound grants that expire automatically, continuously inventory who holds what, and revoke or flag any access that lingers unused, logging grants, expiries and revocations as audit evidence. No statistical formula is supplied; compliance is policy-based. The threshold is that permissions equal the minimum necessary, that grant time-to-live (TTL) is at most one session, and that there are zero unused standing grants older than 7 days. A breach - an over-broad permission, a grant outliving its session, or an unused standing grant past 7 days - triggers automatic revocation and review, shrinking the attack surface before it can be exploited.