# Machine trust

How AI systems are given authority, with zero trust as concrete boundaries. A research area, not a product.

> **Research area.** This is a research area, not a product. Where it mentions Vraelis Contour, it says what exists in the experiment and what is proposed.

## What we study

Four boundaries a request should cross.

- **Explicit identity:** Who is asking? A named principal from a trusted source, not a self-applied label.
- **Limited permissions:** What may it touch? A named resource and action, not carried to other machines.
- **Approval tied to a change and destination:** Who agreed to what, and where? Exact, expiring, separate from the proposer.
- **Checks at execution boundaries:** Where is it enforced? At the moment of action, by whatever holds access.

## What exists today

Research direction, with no product claim.

Vraelis has no identity, access or authorization product, and makes no zero trust claim.

One narrow boundary exists in the Vraelis Contour experiment: release authorization with separate publisher, deployer and approver roles, and destination-bound approval naming the environment and resource, with a 60 second age limit, checked before serving. One check, not a machine trust system.

## What is still open

- **Disconnected operation:** What a machine does when it cannot reach its approval service.
- **Single use approval:** Consuming an approval exactly once, at the boundary.
- **Two approval domains:** Supplier and operator approvals bound to one release.
- **Revocation:** How fast a withdrawn approval reaches every boundary.

## How it connects to Vraelis Contour

Informs: Separate approval roles, a policy checked where it serves.

On vraelis.com: [Machine trust](https://vraelis.com/research/machine-trust), [Critical infrastructure](https://vraelis.com/applications/critical-infrastructure), [Suppliers and integrators](https://vraelis.com/applications/integrators).

- [Model integrity](https://docs.vraelis.com/model-integrity): Whether the model that runs is the model that was approved.
- [Security evidence](https://docs.vraelis.com/security-evidence): Records a person can read, kept in order, including what could not be confirmed.
- [Adversarial threats](https://docs.vraelis.com/adversarial-threats): Inputs, updates and supply paths built to change what a model does.

---

Vraelis documentation, Research. https://docs.vraelis.com/machine-trust