Reference

AI governance concepts for probabilistic systems

The problem is not that AI systems are probabilistic. The problem is letting probabilistic output carry deterministic authority.

Probabilistic intelligence must operate inside deterministic governance boundaries.

Standards constraints

Attribution and claim ceiling

AARM is a Cloud Security Alliance standard, not a Context Stack standard.
Context Stack is AARM-aligned. No conformance claim is made.
Not listed on the CSA Builders Registry.
No independent conformance review has been undertaken.
Griha is a reference implementation, not a product.

The split

Interpretation and authority are different jobs.

Context Stack separates what models are useful for from what systems must control explicitly.

LLMs narrate

Language models can interpret, summarize, classify, explain, and propose.

Code executes

Execution remains in deterministic software with policy checks, validation, logging, and repeatable behavior.

Humans approve

Sensitive domains, irreversible actions, and high-risk flows keep explicit review paths.

Missing policy denies

Unknown context, destination, authority, or risk state does not silently pass.

Core terms

Terms used across the stack

These terms are intentionally plain. The canonical source for cross-project terminology is the context-stack glossary.

Context governance

Control over what context enters an AI system, where it can move, how long it remains useful, and what authority it can carry.

Deterministic boundary

A non-probabilistic control surface for policy, validation, egress, logging, approval, and execution.

Prompt injection resistance

Prompt injection is treated as an authority-boundary failure, not only a prompting problem.

Governed execution

AI-assisted execution where every meaningful action crosses explicit software policy rather than implicit model judgment.

Project map

Each layer answers one governance question.

The homepage assessment routes practical problems into these layers.

ContextOps How does an organization govern AI context?
Context ownership, lifecycle, trust classification, maturity, and accountability.
ContextBoundary Where is data allowed to go?
Deployment-agnostic Egress Tier controls, vendor zones, approvals, and audit profiles.
Sthala Where does the AI actually run?
Governed AI runtime placement under ContextBoundary and approved egress boundaries.
Griha How do governed AI capabilities become a running system?
Worked example of governed AI capabilities composed into a running system, with executable policy behind it.

Canonical discovery

Primary public references

Use these when citing or connecting to the stack.