OPEN DRAFT · V0.1

Permissions for machines that act in the real world.

KineGrant is an open authorization and accountability layer for physical AI. It turns human, object, and space policies into short-lived capabilities an actuator can verify before it moves.

Enter Machine Permission TestRead KGP-001 ↗
Default denyDeny overridesEd25519 signedNo token
KINEGRANT / ACTION SESSION● VERIFIED
ACTION REQUEST
agent: delivery-robot-1target: space-7 / dooraction: openpurpose: delivery
01RequestVALID
02PolicyVALID
03CapabilityVALID
04GateALLOW
05ReceiptVALID
CAPABILITY TTL00:30REPLAY1× ONLY

OPEN SECURITY EVENT / V0.1

Intelligence is not permission.

Try to break replay protection, request binding, expiry, and receipt integrity inside the authorized local sandbox.

Machine Permission Test →

01 / THE MISSING LAYER

Physical AI can act.
It cannot prove it was allowed.

Robot stacks describe what machines can do. Privacy standards describe how data may be used. Device standards control connected endpoints. KineGrant connects them at the decisive moment: immediately before a physical action.

02 / ARCHITECTURE

One narrow core.
Standards at the boundary.

01Request

Agent declares target, action, and purpose.

02Policy

All applicable rules resolve deterministically.

03Capability

Authority signs a short-lived, bound permission.

04Gate

The actuator verifies locally and fails closed.

05Receipt

Executor returns a privacy-minimized proof.

03 / INTEROPERABILITY

Integrate. Do not replace.

KineGrant preserves each standard's authority and adds a common pre-action capability plus post-action receipt.

W3C ODRL

permissions · prohibitions · duties

BOUNDARY ADAPTER

IEEE 7012

machine-readable privacy terms

BOUNDARY ADAPTER

W3C WoT

thing identity · action affordances

BOUNDARY ADAPTER

ROS 2

robot identity · action requests

BOUNDARY ADAPTER

OPC UA

industrial nodes · role permissions

BOUNDARY ADAPTER

Matter

fabric identity · device commands

BOUNDARY ADAPTER

Reference profiles only. KineGrant v0.1 does not claim certification or endorsement by these organizations.

04 / LIVE ACTION GATE

Same robot. Same target.
Different authority.

CHOOSE REQUEST

Simulation of the v0.1 reference policy engine.

kinegrant-gate / local•••
REQUESTdelivery-robot → door-7.openPURPOSEdelivery
  1. 01

    Identity verified

    ✓
  2. 02

    ODRL permission matched

    ✓
  3. 03

    No prohibition matched

    ✓
  4. 04

    Capability issued

    ✓
  5. 05

    Actuator gate opened

    ✓
  6. 06

    Receipt signed

    ✓
DECISIONALLOW

Signed capability valid for 30 seconds and one execution.

05 / KGP-001

A protocol that fails closed.

01

Deterministic

The same request and policy set produce the same decision.

02

Bound

A capability is valid only for one agent, target, action, purpose, and request.

03

Short-lived

Capabilities expire in no more than five minutes and default to one use.

04

Local

Actuator authorization never waits for a public ledger.

05

Auditable

Signed receipts form a tamper-evident chain without raw private media.

Read the online specificationDownload whitepaper · PDF

REQUEST FOR COMMENT

Help define how machines ask before they act.

KineGrant is an experimental draft. We are looking for robotics engineers, security researchers, standards contributors, and independent implementers.

Challenge the draftDownload source · ZIPReference implementation v0.1 · Apache-2.0