SUPEREGO
SUPEREGO Authority Lab

Agents can act.
They still need authority.

SUPEREGO holds consequential actions at runtime until an external authority grants exact, bounded permission. See the boundary work, step by step.

Run the healthcare demo See the control chain
Synthetic dataReal authority pathOne-use permitsVerifiable receipts
01—05

Five views. One authority boundary.

Each scenario isolates a consequential moment where capability alone is not enough.

Scenario 01 · HealthcareLive

Ask before it sends.

A clinic agent assembles a synthetic MRI coverage request, then waits for the ordering clinician to authorize submission to a mock health plan.

  • Exact clinician review
  • One-use permit
  • Replay denial
Open interactive demo
Scenario 02 · FintechLive

Ask before money moves.

A treasury agent prepares a vendor payment, then stops before releasing funds outside the company.

  • Payee binding
  • Amount ceiling
  • Dual approval
Open interactive demo
Scenario 03 · DeveloperLive

Ask before code ships.

A release agent has a synthetic pull request with passing checks, then stops before merging its exact commit and releasing to a mock production environment.

  • Commit binding
  • Release-manager approval
  • Mock GitHub only
Open interactive demo
Scenario 04 · Policy CompilerInteractive lab

Connect tools. Compile authority.

Choose familiar integrations and policy sources, then watch SUPEREGO resolve them into enforceable effects, evidence duties, process steps, connector gaps, and legal review.

  • Granola, Slack, GitHub, MCP
  • NIST, EU AI Act, OWASP
  • Draft is not authority
Open compiler lab
Scenario 05 · Adversarial ContainmentDeterministic lab

Let the planner fail. Contain the action.

A synthetic hostile instruction influences a compromised planner, then real isolated authority controls deny unavailable capability, missing approval, changed subject, replay, drift, and revocation.

  • Compromised-planner fixture
  • Real fail-closed denials
  • Zero protected executions
Open containment lab
THE CONTROL CHAIN

Capability is not authority.

The agent may know how to perform the action. SUPEREGO decides whether this action, for this subject, at this moment, may proceed.

  1. 01
    Agent requests

    The exact action and subject are hashed and held.

  2. 02
    Authority decides

    A person or governed authority reviews the bounded request.

  3. 03
    Permit constrains

    Permission names the action, scope, expiry, and use count.

  4. 04
    Receipt proves

    Execution is matched to authority without exposing content.