Layer 1 — The foundational pillar: Safety
This is the non-negotiable foundation. Safety must be designed into the framework and assured before any other capability is evaluated.
From algorithm to action: A framework for safe and secure physical AI
Key considerations for a new era of autonomous systems
Artificial intelligence (AI) is advancing from digital assistance to physical presence. For decades, AI has operated behind the scenes, optimizing supply chains, producing content, and executing financial trades in various enterprise settings. While significant, much of this impact has been largely virtual. Today, many AI-powered systems can sense, interpret, and physically interact with our world, as well as take control of the everyday machinery that defines modern life.
From autonomous vehicles and robotaxis navigating city streets, to collaborative robots on factory floors, automated data centers, smart energy grids and cities, and a host of other systems, physical AI (PAI) continues to reshape the world around us. This rapid proliferation of PAI across both industrial settings and daily life requires companies to redefine what safety, risk, and trust mean when implementing and monitoring these systems.
AI's next challenge
In the case of an AI-based financial algorithm, its immediate effects and direct outputs are confined to the virtual realm. But a flawed algorithm or compromised PAI system can lead to vehicles swerving into traffic, factory robots dropping heavy objects, or an impaired AI-managed data center triggering a cascading failure that disrupts those relying on its services.
These life-or-death scenarios elevate safety from one of many considerations to the single most critical and immediate imperative. Since a breach in a PAI system can culminate in a safety failure, this demands a holistic approach to security.
The three parts of the problem
Effective management of PAI begins with understanding it as a complete, integrated system. To address this challenge, any robust approach must deconstruct the problem into three interconnected domains, each with its own unique failure modes:
01
The brain (the AI model)
This is the complex decision-making component that processes sensor data and decides what to do. Failures here are failures of intelligence: For example, a model may misidentify an obstacle and decide not to stop.
02
The nervous system (the control system)
This is the signal transmission layer that translates the brain's commands into electronic signals for the hardware. Failures here are failures of execution, such as the brain correctly deciding to stop but a faulty wire preventing the signal from reaching the brakes.
03
The physical body (the asset and ecosystem)
This is the hardware itself—the motors, actuators, and joints, and the environment in which it operates. Failures here are mechanical or environmental. For instance, the brain and control system may work perfectly, but the brake pads are worn and fail to stop the vehicle.
In conversation with KPMG
Understanding the value on the table
1
From pilots to production
Physical AI is moving from pilots to production. The organizations that treat trust as an enabler rather than a checkpoint are the ones that will scale it.
2
Protecting value
Protect the value being unlocked. A single physical incident can erase the operational gains and the reputation behind them. Getting this right protects both the upside and the license to keep operating.
3
Cautious deployment
Deploy PAI with adequate caution and confidence. A clear view of what is safe, who is accountable, and where the limits sit allows an organization to move faster because the hard questions are answered before they become roadblocks.
4
Scalable Frameworks
Build once, scale everywhere. A common framework across the fleet, the supplier base, and every site removes the need to re-litigate trust with each new deployment.
5
Commercial Advantage
Turn trust into a commercial advantage. Buyers now arrive with mature security questions. Being able to answer them with evidence shortens sales cycles and wins the deals where trust is the deciding factor.
From algorithm to action: A framework for safe and secure physical AI
Physical AI is moving from pilots to production. Deploy PAI safely and confidently by establishing clear operational boundaries, accountability, safety and security protocols. Download our report to implement this hierarchical framework and move your organization into the future.
How KPMG can help
- Succeeding with PAI means managing a complete system: a brain (the AI model), a nervous system (the control architecture), and a physical body (the asset and its environment).
- The ability to provide verifiable evidence of safety at every stage is what can unlock regulatory approval, public trust, and ultimately, market leadership.
- This demands a deliberate set of actions: Manage the risks unique to each domain, establish clear responsibility models that close the maker-to-adopter gap before an incident exposes it, and converge learning models with established industrial safety standards.
This is where KPMG is built to help: By combining our market-leading Trusted AI framework with deep industrial technology, strategy, and OT/IoT security capabilities, we help organizations design resilient systems, embed governance, and scale Physical AI safely, securely, and responsibly from strategy through deployment.
Drive innovation
Our results-driven, pragmatic approach to innovation bridges the gap between ideas and execution, empowering you to innovate faster, think bolder, and aim higher.
Cyber security services
As a leading provider and implementer of cyber security, KPMG knows how to apply leading security practices — and build new ones that are fit for purpose.
Technology and digital transformation
Technology alone isn’t enough. It’s the intersection of the right technology, strategic implementation, and continuous innovation that delivers modern technology solutions with real business impact.
Explore related insights
Contact Us
Vijay Subramanyam
Principal, Advisory Technology, Media, and Telecommunications (TMT) Consulting Industry Leader, KPMG US
Brad Raiford
Managing Director, Industrial Manufacturing, Cyber & Tech Risk, KPMG US
Brian Higgins
US and Americas Sector Leader, Industrial Manufacturing, KPMG LLP
Nana Amonoo Neizer
Director, Cyber and Technology Risk, KPMG LLP
Lisa Rawls
Principal, Advisory, Technology, Media & Telecom, Cyber & Tech Risk, KPMG LLP
Vijay Subramanyam
Principal, Advisory Technology, Media, and Telecommunications (TMT) Consulting Industry Leader, KPMG US
Brad Raiford
Managing Director, Industrial Manufacturing, Cyber & Tech Risk, KPMG US
Brian Higgins
US and Americas Sector Leader, Industrial Manufacturing, KPMG LLP
Nana Amonoo Neizer
Director, Cyber and Technology Risk, KPMG LLP
Lisa Rawls
Principal, Advisory, Technology, Media & Telecom, Cyber & Tech Risk, KPMG LLP