Epic: AI-based Access Control (AIAC) — end-to-end delivery

[oblinder]

Summary

AIAC is a Kagenti platform extension that automates access-control policy for AI agents running
on Kubernetes. Instead of hand-writing and maintaining authorization rules, an AI agent
continuously translates a natural-language access-control policy into concrete, enforceable rules —
eliminating manual policy administration and preventing policy drift as services and roles evolve.

This epic is the umbrella for the full AIAC deliverable. It is broken into a documentation track and
four implementation phases; each phase is a deliverable in its own right, and each carries a
runnable demo that showcases the capability it adds.

---

Problem Description

Kagenti AI agents call services across a shared platform. Every call must carry a token scoped to
exactly the permissions the caller's role allows on the target service. Without a dedicated policy
management layer, access policy ends up scattered across per-deployment configuration, creating
three compounding problems:

1. Policy drift — new services and roles are onboarded without matching permission updates,
   because there is no automated mechanism to apply them.
2. Distributed policy intent — there is no single authoritative statement of what each role may
   do; policy knowledge is fragmented across deployments.
3. Manual administration overhead — keeping the rules consistent with a growing fleet of agents
   and tools takes ongoing human attention, with no audit trail.

---

Key Use Cases

• UC-1 · Service Onboarding — when a new agent or tool joins the platform, AIAC discovers what
  it can do and who should be allowed to use it, and generates the access rules automatically. No
  human writes policy by hand. (Delivered progressively across Phases 1–3.)
• UC-2 · Policy Update — when the written access-control policy changes, AIAC re-derives the
  affected rules so enforcement always reflects the current intent. (Phase 4.)
• UC-3 · Role Update — when a role changes in the identity provider, AIAC recomputes every
  policy that role touches. (Phase 4.)

---

Policy Source

The access-control policy starts life as a single written document that AIAC reads in full.
Later, that document is replaced by a searchable knowledge base, so the policy can be updated and
queried live without redeploying a file. The access decisions AIAC produces are the same either way;
only where it reads the policy from changes.

---

Phases (sub-issues)

Tracked as GitHub native sub-issues of this epic. Issue numbers are assigned at post time; the
placeholders below map each draft to its future issue reference.

Sub-issue	Goal	Ref
Documentation	Produce the specifications and the Architectural Summary that describe the end goal	#644
Phase 1 — Functional AIAC Agent + onboarding demo	Stand up the AIAC agent and demonstrate UC-1 onboarding end-to-end, validated by evaluating the generated rules	#645
Phase 2 — Platform integration	Drive onboarding from the live platform trigger path and have enforcement consume the generated rules	#646
Phase 3 — Knowledge-base-backed onboarding	Replace the single policy document with a searchable knowledge base behind the same UC-1 flow	#647
Phase 4 — Policy & Role Update reconciliation	Add UC-2 (policy update) and UC-3 (role update) on top of the knowledge base	planned — Round 2