The Human Layer
for the
Autonomous Internet.
Permission Zero is a layered authority and defense stack: prove a human root, identify autonomous actors, delegate exact permissions, evaluate policy, then contain swarm-scale abuse.
Every action is evaluated
before authority becomes execution.
This model shows how Permission Zero turns a human decision into bounded machine authority. An agent does not receive permanent trust simply because it has an identity. Every meaningful request is evaluated against the authority that was actually delegated.
The diagram is an architectural model for the Genesis specification; it is not a claim of a production deployment.
POLICY ENGINEidentity · scope · behavior · contextFive responsibilities.
One authority model.
Permission Zero separates proof, identity, delegation, policy and swarm defense because each solves a different trust problem. Keeping the boundaries explicit makes the system easier to reason about, test and evolve.
Human Proof. Establish that an accepted human root exists while revealing only the attributes a relying service actually requires.
Agent Authority. Give autonomous actors distinct identities and bind those identities to the human or organization that authorized them.
Delegation. Express authority as explicit capabilities: action, resource, amount, time, risk and revocation boundaries.
Policy Engine. Evaluate meaningful actions at decision time using scope, context, behavior and current risk.
Swarm Defense. Look beyond one request and detect coordinated abuse across identities, endpoints and time.
Define what an agent can do—nothing more.
Illustrative policy behavior for the Genesis specification.
| Example action | Low-risk agent | High-risk agent | Unverified agent | Hostile swarm |
|---|---|---|---|---|
| Read public data | Allowed | Allowed | Rate limited | Blocked |
| Access user profile | Allowed | Rate limited | Blocked | Blocked |
| Initiate payment | Approval | Blocked | Blocked | Blocked |
| Transfer funds | Approval | Blocked | Blocked | Blocked |
| Post content | Allowed | Rate limited | Blocked | Blocked |
| Create credentials | Approval | Blocked | Blocked | Blocked |
One authority model.
Different operating environments.
The same authority model can sit behind very different applications. The policy changes by environment, but the core questions remain consistent: who authorized the action, what exactly was delegated, is the action still inside scope, and does the surrounding behavior look coordinated or abusive?
Commerce. Allow customers and approved shopping agents to transact while containing account abuse, inventory capture and promotion exploitation.
Banking & FinTech. Separate research and preparation from money-moving authority, with explicit human approval for sensitive actions.
Enterprise & developer platforms. Bound autonomous access to internal data, APIs, infrastructure and administrative operations.
Social and AI applications. Distinguish human-rooted agents from unknown automation and coordinated manipulation without requiring blanket identity exposure.