What is Physical AI on Intended

How the Intended Authority Runtime layers on top of an existing safety stack to gate AI-issued commands at machine speed, with a tamper-evident audit trail.

beginner4 min readimplemented

What is Physical AI on Intended#

Intended is the authority layer for physical AI. When agentic robots, drones, AGVs, or autonomous machines are about to take a real-world action, Intended classifies the intent, mints a short-lived signed credential bound to the robot's identity, and falls back to a declared safe-default if the credential is denied or expires.

We sit above your existing safety PLC, not in place of it.

The picture#

Operator HMI / autonomous planner / AI agent
            │
            ▼
   ┌──────────────────┐
   │  Intended layer  │   ← classify intent, issue token
   │ (this is new)    │
   └────────┬─────────┘
            │ token + safe-default
            ▼
┌────────────────────────────────────────────────────┐
│  Existing safety PLC + bus (unchanged)             │
│  e-stop chain, light curtains, presence scanners   │
└────────────────────────────────────────────────────┘
            │
            ▼
       motion controller
            │
            ▼
          motors

The safety PLC continues to be the authority on physical safety hazards. Intended is the authority on whether the command should be issued in the first place.

What you get#

	What it is	Where it lives
OIL v2 — Open Intent Layer	Open Apache-2.0 taxonomy of 173 categories across 29 domains (15 physical / embodied + 14 digital). The vocabulary your robot uses to describe what it's about to do.	github.com/intended-so/intended/tree/main/packages/open-intent-layer
Authority Runtime — cloud	Classifier, policy evaluator, signing issuer, audit chain. RS256 JWTs verifiable against published JWKS.	`https://api.intended.so/v1/physical/*`
Edge verifier — Rust	Sub-50ms local verification. JWKS offline cache (24h). Revocation list sync. Pre-computed authority cache.	github.com/intended-so/intended/tree/main/crates/intended-verifier
SDKs	Python, TypeScript, ROS2 ament. C++ and Rust customer-facing SDKs ship with first design partner.	github.com/intended-so/intended/tree/main/packages
Reference cobot	"Hello world" of physical-AI authority — ROS2 / MoveIt-shaped pick-and-place that runs on a laptop without a robot.	github.com/intended-so/intended/tree/main/examples/physical-ai/pick-and-place
Safety case template	GSN sketch of how to extend our certified primitive into your machine's safety case.	Safety-case writing

What we do NOT do#

Replace your safety PLC, safety bus, or safety-rated I/O
Write controls code, ROS2 nodes, ladder logic, or PLC programs
Read your PLC tags directly — you implement a PhysicalStateProvider
Provide platform-specific adapters for ABB, FANUC, KUKA, UR, Boston Dynamics
Decide where to grasp, what trajectory to plan, or which lane to merge into
Currently certify under IEC 61508 / ISO 13849 / ISO 26262 — sales-funded when regulatory posture demands it. The primitive is buildable today; un-assessed safety-relevant SW is a defensible posture under your own SW lifecycle (the Linux-upstream model).

Where we fit in your safety case#

Customers ship a safety case for their machine. We provide a certified primitive their case can cite. The split:

Component	Owner
Edge verifier binary	Intended
Cloud authority issuer	Intended
Robot / PLC / safety bus	Customer
Application logic + DAG	Customer
Safety bus → predicate bridge (PhysicalStateProvider)	Customer
Per-deployment safety case	Customer

Safety-case writing walks through the GSN structure and what's defensible to claim.

Two-product framing#

Critical distinction we never collapse:

Product	Open / closed	Pricing
OIL — the standard	Open, Apache 2.0, vendor-neutral	Free forever
Intended Authority Runtime + edge verifier	Commercial product	Per-device or per-cell, annual

You can adopt OIL without buying our runtime. You buy our runtime when you want OIL enforced at machine speed with the safety, audit, and operational properties enterprises require. Same shape as Linux + Red Hat, Postgres + RDS, OAuth + Okta.

Next#

Physical-AI quickstart — five minutes to first token
ROS2 integration guide — canonical SDK for robotics agents
PLC migration patterns — translating safety-PLC patterns into Intended policy clauses
Safety-case writing — what to claim, what not to claim
OIL v2 — the open standard — vocabulary your robot uses to describe what it's about to do