EXECUTABLE ASSURANCE

Films and explainers · Ernest Chiweshe

Videos

Start here, then the architecture, then the reference.

Each video plays from YouTube only when you choose it. The papers and the reference code are linked below.

Playlist 1

Start here

One Door: Executable Assurance

15:24

Alongside all the excitement about AI, nearly every thoughtful person carries a quiet, unspoken fear. This film names it, and answers it with one idea: capability is not authority.

One Door: Capability Is Not Authority | Official Trailer

2:58

The three-minute introduction to One Door. An AI can think all it likes; the right to act on our world is something we grant, one action at a time, through one door that starts closed.

Who Holds the Keys? AI Capability Is Not Authority

8:32

AI safety and AI security ask different questions. Learn how to tell them apart, why an AI agent’s capabilities do not grant it authority, and how to read the next AI headline.

Playlist 2

The architecture

The Model Is Not the Boundary: Governing AI Actions

21:05

An architecture briefing for security and engineering leaders on governing actions outside the model.

Where Everything Lives: An AI Authority Architecture

21:46

An architectural map for developers, security engineers and deployment teams.

NVIDIA OpenShell Explained: The Fence Outside the Model

18:50

On 21 October 2026 NVIDIA published the OpenShell reference architecture: agents think on Rubin, act on Vera, and are governed on BlueField.

The Approval Is Not a Click: What Makes a Human Yes Real

12:39

An approval is a signature on an exact decision, by a verified person, inside a window, recorded before anything moves. Anything less is a click.

Playlist 3

The reference

One Door: Why Capability Is Not Authority in AI

6:18

An AI agent that can act is not the same as an AI agent that may act.

One Door, for Reviewers: The Authority Path in Detail

6:55

The same door, for people who will check it: request and principal, the policy decision, a durable clock, evidence and retention, witnesses, re-decision, the single-use permit, the receiver, adversarial cases, and the cost and limits.

Capability Is Not Authority: The Architecture

9:17

How a network engineer broke the "agentic operating system" into layers: a core that knows nothing about any industry, three runtimes each wrapping the next, domain packs, and one fail-closed authority path for every consequential action.

Five Signed Lines: Inside Executable Assurance

24:48

A file-by-file walkthrough of the reference implementation: how a decision becomes five signed lines of evidence before any effect is released.

Published 22 October 2026

The papers and the code