Catalog of SIP Cores
System on Chip design resources
System on Chip design resources
In part two of our blog series on securing physical AI, Nicole Kostopoulos continues to explore what it takes to build security into physical AI products across their full lifecycle. If you missed part one, you can catch-up via this link where we discuss the shift from cloud AI to Physical AI and why trust, embedded security, and new operating models will define how AI is deployed in the real world.
August 3, 2026 -
The industry consensus is that trust will define the next era of innovation as AI moves beyond the data centre and into the physical world. But if trust is the defining requirement of the physical AI era, the next question is how to achieve it. As AI systems move from digital environments to real-world operations, the industry must adopt new approaches to security, resilience and lifecycle management; all of which must be able to keep pace with rapidly evolving threats.
Beyond the technology itself, during his recent GSA Executive Forum keynote, Christophe Nicolas, SVP of Kudelski Labs, highlighted a growing operational challenge as our industry starts to consider the challenges posed by physical edge AI as solutions move from concept to deployment.
The rapid growth of AI in general is accelerating how vulnerabilities are discovered, significantly increasing both the volume and pace of critical issues. As Christophe referenced during the keynote, the industry is entering its “Mythos moment”, (a reference to the launch of the latest Anthropic Claude model, which has provided a step-change in autonomous vulnerability discovery and exploitation). This means AI is not only improving systems, but also dramatically increasing the rate at which weaknesses are found.
What used to be manageable through periodic updates is quickly becoming unsustainable.
As Christophe pointed out, if every system requires constant patching to remain secure, the model starts to break down. Physical assets such as vehicles or industrial systems cannot simply stop and update continuously without impacting their operation.
This raises a fundamental question: are existing security and update models still fit for purpose in a world of autonomous, real-time systems?
To adapt, organizations need to:
In this context, the transition to physical AI is not only a technology shift, it requires a fundamental change in how systems are designed, secured, and operated.
Joining Christophe at this year’s GSA Executive Forum was KLARQ, the Kudelski Labs Autonomous Robotics Quadruped. Demonstrating the application of security to physical AI, KLARQ showed how Edge AI, embedded security, a trusted device and AI-based inference can be integrated with an existing, off-the-shelf robot.
KLARQ demonstrated what happens when AI leaves the data centre and enters the physical world, where decisions have real-world consequences and trust becomes a design requirement rather than a feature. KLARQ showed the audience how autonomous systems can securely infer, decide and act while maintaining integrity through the device lifecycle.
Providing a tangible example of the central message of Christophe’s presentation, KLARQ showed that the organizations that succeed will not simply deploy more intelligent systems but systems that are trusted, controlled and resilient by design. Physical AI is not only about deploying capable models but deciding where critical decisions are made, how those decisions are protected and how the system remains trustworthy in real-world conditions.