BLACK OCTOBER, PART 2 OF 7 — CONSIDERATION
By Mark Rogge, CEO of EnforceAuth | 8 min read | for security architects and platform engineering leaders evaluating authorization platforms
If Part 1 convinced you that Black October's incidents were an authorization problem, not a safety problem, the next question is harder: how do you actually evaluate a policy-as-code authorization platform without getting sold a feature list that doesn't survive contact with your environment?
After watching this category get evaluated — and mis-evaluated — across dozens of enterprise deployments, here's the framework we'd want if we were on the buying side.
1. Does it cover all four domains, or just one?
AI agents don't respect your org chart. A single agent action might touch an application API, pull from a database, and trigger an infrastructure change in one session. If your authorization platform only covers one of those domains — say, API gateways — you've secured a quarter of the attack surface and left the rest to whatever was there before. Ask any vendor directly: does this one policy engine span applications, infrastructure, data, and AI workloads, or do I need three more tools to cover the rest? The honest vendors will tell you where their coverage stops.
2. Is it built on an open standard, or a proprietary engine?
Policy-as-code only delivers on its promise — versioned, testable, portable authorization logic — if the policies themselves aren't locked to one vendor's syntax. OPA/Rego has become the closest thing this space has to a common language: battle-tested, widely adopted, and not owned by any single vendor. A platform built natively on OPA means your policies are an asset you own. A platform built on a proprietary rules engine means your policies are a migration project waiting to happen the day you want to leave.
3. Does it enforce at runtime, or just audit after the fact?
This is the distinction that separates a monitoring tool from an authorization platform. A system that logs what an agent did and flags it for review is useful for forensics. It does nothing to stop the bad transaction, the cross-tenant query, or the production push in the moment it happens. Runtime enforcement means the policy decision happens *before* the action completes — deny by default, allow by policy — not after someone reviews a dashboard.
4. Can it handle both human and non-human identity, continuously?
Most identity and access tooling was built to authenticate a human once at login and trust the session from there. AI agents don't work that way — an agent's context, task, and risk profile can change mid-session, and a policy that was correct at the start of a workflow can be wrong three steps later. Continuous identity verification — not just a one-time login check — is the difference between a platform designed for humans and one designed for the 82:1 reality most enterprises are actually living in.
5. What does it give you when someone asks "why did this happen?"
This is the question every platform answers badly until the day you need it answered well. A board member, a regulator, or an incident commander is going to ask which policy version made which decision, for which identity, and why. If the answer requires stitching together logs from three systems and hoping the timestamps line up, you don't have an authorization platform — you have a liability. This is exactly the gap EvidenceIQ was built to close inside EnforceAuth: a continuously maintained context graph of every identity, every relationship, and every authorization decision, so that evidence is a query away, not a project.
6. Does the commercial model survive your actual growth, not just your pilot?
Authorization decisions scale with usage, not headcount — a successful AI initiative can 10x your decision volume in a quarter. Ask how pricing behaves at that inflection point before you're locked into a number that made sense in the proof-of-concept and stopped making sense the moment the project succeeded.
The buying mistake underneath all six
Every one of these questions comes back to the same test: does this platform treat AI agents as a bolt-on to human identity management, or as a first-class identity type from day one? Most of what gets marketed into this category was retrofitted — application-era IAM with an "AI" label added to the brochure. The platforms that will still be defensible in two years were built assuming non-human identities were always going to outnumber human ones.
Part 3 of this series goes one layer deeper: what "best" actually means once you've run this framework, and why the honest answer depends more on your architecture than on any vendor's marketing page.
Want to run this six-question framework against your own shortlist? Reach out at enforceauth.com and we'll do it with you, vendor-neutral.
EnforceAuth is the AI Security Fabric — policy-as-code authorization built natively on OPA/Rego, covering applications, infrastructure, data, and AI workloads, with EvidenceIQ providing continuous audit-ready evidence on every decision.
About EnforceAuth
EnforceAuth is the OPA/Rego authorization control plane. The customer's OPA decides what people, services, and AI agents may do at action time. Verdict covers high-impact agent actions. EvidenceIQ keeps the record behind every decision.
