QenSAi · AIAGF · Governance in UseWalk guide · updated 14 October 2026
How to walk this site

Start with the story.
Branch where you want to see it run.

One route through the public material for people and for machines. Follow the four steps in order, or leave the story at any stage to see how that stage runs in operations, then come back.

Start with the story: One Request, Governed →Prefer pictures? The visual guideMachine-readable routes (index.json)

Where to begin

The reading order

  1. Step 1 · One Request, Governed
    The plain-language story: nine stages from a person's request to a human decision, then the recorded runs.
  2. Step 2 · Engineering evidence review
    Each recorded case with its supported conclusion, review question and limits.
  3. Step 3 · The full account
    The report, with process evidence, the evidence annex and the scope-correction case.
  4. Step 4 · Machine-readable records
    index.json (reading order and stage routes), the JSON evidence files, SHA256SUMS.txt and READ-ME.txt.

The nine stages, and where each one branches

Each stage of the story ends with a "See it in operations" line. It opens the matching step in the public view of the fifteen operations steps. Operator pages need a sign-in; the public view does not.

Story stageOperations steps (public view)More to read
1 · A1 RequestStep 8: Find the request as staged work—
2 · Front gateStep 11: A person authorizes the work
Step 12: Only accepted work runs
How a card becomes a governed task (operator diagram)
3 · Oversight record (SPOC form)Step 1: Start at the Response Table
Step 3: Review the oversight record
Step 4: Find the gap that is already recorded
How the oversight form is graded (operator diagram)
4 · Scope (NTF)Step 2: Find the rules that apply
Step 6: Tie each measure to a real rule
How a new frame is added (operator diagram)
5 · Thread (2VP)Step 7: Compose or reuse the thread
Step 9: Attach the thread to the task
How the 2VP links screens to records (operator diagram)
6 · ExecutionStep 10: Prove where the work goes
Step 12: Only accepted work runs
—
7 · Truth gateStep 13: Prove the result, not just the reply
Step 14: Trace one chain of evidence
—
8 · Response TableStep 1: Start at the Response Table
Step 15: Return it to the record that governed it
Response Table — annotated guide
9 · Record and returnStep 13: Prove the result, not just the reply
Step 15: Return it to the record that governed it
From a human request to a governed result · Signal Capture

Side journeys

Mapped to the OWASP Top 10 for Agentic Applications

Where each risk in the OWASP list is addressed in the story, and the evidence today.

RiskStory stagesEvidence today
ASI01 · Agent Goal Hijack1, 3, 4, 7Recorded: an approved correction changed the next response (15 Sep 2026); a person redirected an agent that had made an unfinished experiment its goal, and the correction was kept in the oversight record (9 Oct 2026, record-keeping only).
Open cited evidence · recorded
ASI02 · Tool Misuse & Exploitation4, 5, 6Designed. Tool and MCP server hardening is not shown.
Open cited evidence · partial
ASI03 · Identity & Privilege Abuse2Recorded: a person authorized before anything ran (15 Sep 2026). Workload-identity standards such as SPIFFE or OIDC for agents are not shown.
Open cited evidence · recorded
ASI04 · Agentic Supply Chain Vulnerabilities4, 9Designed. Third-party packages and registries are outside what we show.
Open cited evidence · partial
ASI05 · Unexpected Code Execution (RCE)2, 6Designed for authorization. Sandboxing and network isolation are outside our layer.
Open cited evidence · partial
ASI06 · Memory & Context Poisoning3, 4Recorded in part: the task contract and its records now travel in the agent's first message (2 Oct 2026).
Open cited evidence · recorded_in_part
ASI07 · Insecure Inter-Agent Communication6, 8Recorded in part: one desk carried one request to three specialist agents, with a person arming each request (2 Oct 2026). Detecting unsanctioned side channels is not shown.
Open cited evidence · recorded_in_part
ASI08 · Cascading Failures7Recorded: the gate caught a false "done" and stopped the chain, then exposed its own gap when it counted an honest "no data" reply as completed (1 Oct 2026, two local runs); the gap was fixed on 2 Oct 2026.
Open cited evidence · recorded
ASI09 · Human-Agent Trust Exploitation7, 8Recorded: completion is now checked against criteria; a claim with no evidence fails (2 Oct 2026).
Open cited evidence · recorded
ASI10 · Rogue Agents2, 7Designed. Containment of an agent that has escaped its environment is outside our layer.
Open cited evidence · partial

Outside our layer: Sandboxing and network egress control; Protection of binaries against reverse engineering; Containment of agents that have escaped their environment.

What needs a sign-in

Put the governance claims under pressure.

The Governed Escape Test preserves its preregistered conditions and now exposes a bounded partial result with its limits. This optional advanced handoff comes after the story, proof and maps; it is not another starting point.

Enter the Governed Escape Test →I am still exploring

Each page labels its own evidence: recorded proof, designed, illustration, live read view or explanation. A label tells you what kind of claim the page makes; read the limits each page states.

← Back to the story · See the Proof · Qensai