Roles and Access (RBAC / ABAC) — Guardrails and Fix Pattern

🧭 Quick Return to Map

You are in a sub-page of Governance.
To reorient, go back here:

Governance — policy enforcement and compliance controls

WFGY Global Fix Map — main Emergency Room, 300+ structured fixes

WFGY Problem Map 1.0 — 16 reproducible failure modes

Think of this page as a desk within a ward.
If you need the full triage and all prescriptions, return to the Emergency Room lobby.

This page defines role-based access control (RBAC) and attribute-based access control (ABAC) guardrails for AI pipelines.
Without explicit access boundaries, LLMs may read from unintended sources, leak sensitive data, or bypass audit policy.

When to use this page

Your RAG or agent stack integrates multiple data stores with different sensitivity levels.
You cannot trace who accessed what across prompts, embeddings, or tool calls.
Evaluation runs fail because different users see different knowledge bases.
Compliance requires proof of least privilege but no policy schema exists.

Acceptance targets

100% of RAG data calls tagged with role or attribute context.
Coverage ≥ 0.95 of sensitive datasets behind access boundaries.
Audit trails record who, what, when, ΔS, λ_state.
Role drift probes show λ remains convergent across 3 paraphrases.
Exceptions logged with owner and expiry date.

Common failures → exact fixes

Symptom	Likely cause	Open this
Agents fetch data beyond allowed scope	missing RBAC fences	policy_baseline.md
Two users get different citations	inconsistent ABAC checks	retrieval-traceability.md
Logs don’t show who triggered retrieval	no role injection	data-contracts.md
Role drift causes schema injection	misplaced role attributes	prompt-injection.md
Sensitive snippets leak in chains	missing attribute check	pii_handling_and_minimization.md

Fix in 60 seconds

Attach context
Every retrieval call carries {role, attribute_set, index_hash, ΔS, λ_state}.
Enforce least privilege
Roles map to dataset groups. Attributes refine down (e.g. geography, project).
Log every decision
Audit trail logs query, ΔS, λ state, role, attributes, and snippet ids.
Probe role drift
Run 3 paraphrases per role. If λ flips, enforce schema lock.

Minimal copy-paste checklist

Define roles (admin, annotator, auditor, agent).
Define attributes (region, dataset sensitivity, project scope).
Attach {role, attr} to all tool and retrieval calls.
Enforce least privilege at ingestion and retrieval.
Log ΔS and λ_state by role.
Review and expire waivers.

🔗 Quick-Start Downloads (60 sec)

Tool	Link	3-Step Setup
WFGY 1.0 PDF	Engine Paper	1️⃣ Download · 2️⃣ Upload to your LLM · 3️⃣ Ask “Answer using WFGY + <your question>”
TXT OS (plain-text OS)	TXTOS.txt	1️⃣ Download · 2️⃣ Paste into any LLM chat · 3️⃣ Type “hello world” — OS boots instantly

Explore More

Module	Description	Link
WFGY Core	Canonical framework entry point	View
Problem Map	Diagnostic map and navigation hub	View
Tension Universe Experiments	MVP experiment field	View
Recognition	Where WFGY is referenced or adopted	View
AI Guide	Anti-hallucination reading protocol for tools	View

If this repository helps, starring it improves discovery for other builders.

要我直接繼續幫你生出來嗎？

5 KiB Raw Blame History Unescape Escape