What is an AI governance runtime?

An AI governance runtime is the control plane that enforces policy while AI systems operate, in-line with each action, rather than only in review cycles. EVE AI Core evaluates every proposed action from a model, agent, or tool before it executes and records the decision as signed, replayable evidence.

How is a runtime different from static governance?

Static governance produces documents, attestations, and periodic reviews that describe intended behavior. A governance runtime acts in the request path: it evaluates the actual action before it runs and returns an ALLOW, BLOCK, or MODIFY verdict for every governed decision, with evidence.

Does it govern autonomous agents and tool use?

Yes. Tools and agents must be registered to run, and each step of an autonomous plan is a discrete governed decision with its own risk score, verdict, and evidence. High-risk steps can require human approval before proceeding.

How does it deploy into an existing stack?

As a REST decision API (POST /v1/decisions/evaluate), via Python, TypeScript, and Rust SDKs, or as a sidecar in front of existing services, with webhooks for governance events. Because it governs the decision rather than the model, you can change providers without changing the control plane.

Is it fast enough to sit in the request path?

The decision path is deterministic with no LLM in it, and each decision records its own latency in the evidence. It is designed to run in-line; the CoreGuard demo shows the enforcement pipeline and per-decision latency.

How does this relate to EVE CoreGuard?

EVE CoreGuard is the enforcement engine of the runtime — the fail-closed gate that returns the verdict and emits the signed certificate. This page describes the runtime category; CoreGuard is the product and the live demo.

AI Governance Runtime

The Category

What an AI governance runtime is

An AI governance runtime is the control plane that sits between AI intent and real-world execution. Governance must happen while AI systems are operating — not only during quarterly review cycles or in a policy document. EVE AI Core is that runtime: every proposed action from a model, agent, or tool is evaluated against policy the instant before it would run, unsafe actions are blocked or modified, and every decision is recorded as signed, replayable proof. Static governance describes what should happen; a governance runtime decides what actually does.

Runtime vs. Static

Runtime enforcement vs. static policy documentation

	Static policy & review	EVE AI governance runtime
When it acts	During review cycles, after the fact	In-line, before each action executes
What it controls	Intent, on paper	The actual action, at the gate
Coverage	Sampled and periodic	Every governed decision, every time
Output	Reports and attestations	ALLOW/BLOCK/MODIFY verdict + signed evidence
Agent readiness	Not designed for autonomous tool use	Governs each agent step as a discrete decision

Agents & Tools

Tool-use governance for enterprise agents

Autonomous agents are dangerous precisely because they act — calling tools, moving data, taking steps no one reviewed. A governance runtime treats each tool call as a discrete decision that must clear the gate. Tools and agents must be registered to run at all; every step is risk-scored and evaluated against policy; and high-risk steps can require human approval before proceeding. A multi-step plan is therefore a sequence of individually governed, individually evidenced decisions — not a black box you inspect afterward.

Every governed decision passes the gate before execution; the verdict and its signed evidence are produced at the same instant — including for blocked actions.

Enforce, Then Prove

Blocking before execution, proof after every decision

Block before execution

EVE CoreGuard evaluates every proposed action before it executes and returns a deterministic ALLOW / BLOCK / MODIFY verdict against versioned policy packs — fail-closed, with no LLM in the decision path — nothing unsafe reaches the world unless it clears the gate.

Prove after the decision

Each verdict emits an Ed25519-signed certificate and appends to a hash-chained trail, replayable on demand and verifiable offline.

Model-agnostic

The runtime governs the decision, not the model, so you can change providers or self-host without changing the control plane.

Deploy in-line

Integrate as a REST decision API or as a sidecar in front of existing services — no rearchitecting required.

In Your Stack

How EVE CoreGuard fits into enterprise AI infrastructure

The engine of the runtime is EVE CoreGuard — the fail-closed gate that returns the ALLOW / BLOCK / MODIFY verdict. It sits in the request path via a REST decision API (POST /v1/decisions/evaluate), Python / TypeScript / Rust SDKs, or a sidecar in front of existing services, and emits governance events by webhook. To see the runtime in action — a preloaded blocked scenario, the enforcement pipeline, the signed certificate and replay ID — run the live CoreGuard demo.

Common Questions