The Era of Physical AI: Is Cybersecurity Ready? Extension of the Purdue Model

 

The Era of Physical AI: Is Cybersecurity Ready? Extension of the Purdue Model



1. Why Physical AI Security Matters Now

Industrial environments are rapidly evolving beyond traditional PLC-based automation. Autonomous mobile robots (AMRs), AI-powered inspection systems, and even humanoid robots are entering factories and warehouses.

A representative example is Boston Dynamics and its humanoid robot Atlas.

Have you seen the recent article about Hyundai Motor Group planning to deploy Boston Dynamics robots in future factories? The era of Physical AI is no longer distant, it is arriving now.

These Physical AI systems are no longer simple machines.

They include:

  • AI-driven decision-making

  • Cloud connectivity

  • Over-the-air (OTA) updates

  • Real-time motion control

  • Wireless-first networking

In essence, they are fully realized Cyber-Physical Systems (CPS).




2. CPS Security: Where IT, IoT, IIoT, and OT Converge

2.1 What They Have in Common

Across IT, IoT, IIoT, OT, and Physical AI, we observe shared characteristics:

  1. Network connectivity

  2. Software-defined behavior

  3. Remote management and updates

  4. Identity and cryptographic trust

Core security mechanisms still originate from IT:

  • TLS

  • PKI and certificates

  • Identity & Access Management (IAM)

  • Patch management

  • Logging and monitoring

From a technology stack perspective, Physical AI looks highly “IT-like.”


2.2 The Critical Difference: What Is at Risk?

DomainPrimary Asset at RiskImpact of Compromise
ITDataData breach
IoTDevices / privacyPrivacy loss
IIoTIndustrial telemetryOperational disruption
OTPhysical processesSafety & production incidents
Physical AIAutonomous physical actionImmediate physical harm

Physical AI systems, like OT, directly influence the physical world.

A compromise may not result in stolen data, it may result in:

  • Collision

  • Equipment damage

  • Worker injury

  • Production shutdown

This aligns far more closely with OT risk models than IT.


3. Why Physical AI Is Closer to OT

Despite running Linux, containers, and APIs, Physical AI systems share fundamental OT properties:

  • Deterministic real-time control loops

  • Sensor → control algorithm → actuator architecture

  • Tight coupling with safety mechanisms

  • Availability as the highest priority

Unlike IT systems, downtime is not merely inconvenient, it can be dangerous.

Thus, while the technology stack resembles IT, the operational risk model resembles OT.


4. Viewing Physical AI Through the Purdue Model

The foundation of OT architecture remains the
Purdue Enterprise Reference Architecture.

Traditional Purdue levels:

LevelDescription
Level 0Sensors & actuators
Level 1Controllers (PLC)
Level 2SCADA / HMI
Level 3Site operations
Level 4-5Enterprise IT

4.1 Where Does Physical AI Fit?

Physical AI does not sit neatly at a single level.

Instead, it compresses the vertical Purdue stack into a mobile, distributed system:

Level 0

  • Motors

  • LiDAR

  • Cameras

  • Force sensors

Level 1

  • Real-time embedded controllers

  • Motion control boards

  • RTOS-based control

Level 2

  • Fleet management systems

  • Local supervisory services

Level 3

  • Mission scheduling

  • Integration with MES/ERP

Level 4

  • Cloud analytics

  • AI model training and deployment

A robot is effectively a vertically integrated Purdue architecture on wheels.

That architectural compression fundamentally challenges traditional OT segmentation assumptions.


5. Structural Similarities Across CPS Domains

5.1 Similar to IT

  • Linux-based OS

  • Containerized workloads

  • API-based control

  • Cloud dependency

  • Remote updates

5.2 Similar to OT

  • Real-time constraints

  • Fieldbus/EtherCAT communication

  • Safety loops

  • Availability-first design

  • Change management over aggressive patching

Physical AI sits at the convergence of these two worlds.


6. Emerging Security Challenges in Physical AI

6.1 The Collapse of Static Network Boundaries

Traditional OT segmentation assumes fixed assets.

Robots move.

Zone definitions become dynamic rather than physical.


6.2 Wireless Dependency

Wi-Fi, private LTE, and private 5G become core operational infrastructure.

Security now must account for:

  • Roaming stability

  • Authentication integrity

  • Wireless interference risks

  • Encrypted telemetry inspection challenges


6.3 OTA as Both Security and Risk

Updates improve security posture.

But OTA pipelines also introduce:

  • Supply chain attack risk

  • Compromised update servers

  • Malicious firmware propagation

The update channel becomes a high-value target.


7. Reinterpreting Purdue for the Physical AI Era

To secure large-scale robot fleets, it is increasingly practical to introduce a logical extension to Purdue:

The “Robot Zone” (Conceptually Level 2.5)

This zone typically includes:

  • Wireless segmentation (dedicated SSID or 5G slice)

  • Robot VLANs

  • On-premise fleet manager

  • Controlled cloud egress via industrial DMZ

This approach:

  • Preserves OT segmentation principles

  • Prevents uncontrolled enterprise exposure

  • Maintains operational reliability

It extends, rather than replaces, the Purdue model.


8. Conclusion: CPS Security Starts with OT, But Cannot Ignore IT

Physical AI systems are:

  • IT in technology

  • OT in impact

  • CPS in architecture

Therefore, security strategy must integrate:

  • IT-grade identity, encryption, and monitoring

  • OT-grade availability and safety prioritization

  • Purdue-based zone and conduit modeling

  • Secure wireless and cloud governance

Physical AI is not a temporary trend.

Autonomous robotics, AI inspection, and mobile automation are reshaping industrial environments.

Security architecture must evolve accordingly, not by abandoning OT principles, but by expanding them into a mobile, AI-driven CPS reality.

Comments

Popular posts from this blog

Don't confuse DCS, PLC and SCADA in front of OT specialists

Let's create our own ICS Labs in the VMs!

Ship Security: Scope, Schedule, Penalties, etc. for IACS UR E26 and E27 and other regulations