Authorization must cease to be a static grant and become a dynamic, multi-dimensional decision — an Authorization Matrix that moves with context, not a role list written in stone.
In 2014, Google did something radical: it abolished the corporate intranet. Since then, employees accessing company systems no longer rely on the coarse assumption that "being on the intranet = being trustworthy." Instead, every request is judged dynamically against real-time signals — device health, user behavior, geographic location, and more. In 2020, NIST turned this philosophy into a standard: SP 800-207, Zero Trust Architecture, whose core can be summed up in one sentence: never trust, always verify.
What is striking is that more than six years after the standard's publication, most systems' permission models remain stuck in the ACL and RBAC era. Both work well enough in small-scale systems, but they share the same flaw: permissions are written in stone at the moment of authorization. Once a user is assigned a role, they hold a fixed pass: no matter where they are, what device they use, or what level of data they touch, the pass grants it all. Today, as the number of systems, providers, and partners has exploded, splitting roles ever finer leads to unmanageable complexity, while not splitting them finely enough inevitably leads to over-provisioning. Being static is the fatal flaw both models share.
The way out is not to patch the old models, but to turn authorization from a static "act of granting" into an ongoing "decision process." The complete authorization process consists of two parts:
Authentication (AuthN): confirming "who you are" — verifying identity via passwords, digital signatures, biometrics, device fingerprints, one-time keys, and the like.
Authorization (AuthZ): deciding "what you can do" — based on the runtime context (the application involved, business process, time, region, device state, risk level, etc.) and the confirmed identity, granting the right to access specific resources within a certain time window. Resources here are broadly defined: API endpoints, operation sets (read, write, approve, export), object sets, and even the scope of specific data content.
The gap in the old models becomes clear in comparison. After AuthN, RBAC makes all authorization decisions based on a single dimension — "which role you carry" — whereas complete AuthZ is inherently multi-dimensional: beyond identity, there are dimensions of application, process, time, geography, and resources, all changing in real time with the runtime state. A finance manager exporting a report from an office computer during working hours and the same manager bulk-pulling customer data at 3 a.m. from an unrecognized device should receive entirely different answers. This is precisely what an Authorization Matrix is meant to hold: not a static role list, but a decision table whose values shift dynamically with environmental signals.