Securing Edge AI through trusted computing

In today’s computing landscape, edge computing has proven to be a highly viable and low-latency option for real-time operations. By leveraging local compute devices – such as IoT nodes, small form-factor desktops, and edge servers – applications can execute closer to the source of data, significantly reducing the latency and bandwidth limitations typically associated with centralized cloud or traditional physical infrastructure. Rather than using a centralized model that uses shared resources to host applications, edge computing instead is decentralized, placing resources closer to the users and businesses requiring them.

However, times are changing. The rise of Artificial Intelligence (AI) has changed operations across a number of different industries and sectors, and computing is no different. As a result, businesses are increasingly adopting ‘Edge AI’ in order to enhance their operations.

What is Edge AI?

At a basic level, Edge AI involves executing AI algorithms on edge devices allowing data to be processed in near real time, close to where it is generated. While this reduces the need for constant connectivity, Edge AI can still seamlessly integrate with cloud environments to offload heavier task or aggregate insights at scale.

AI processing is often bandwidth-intensive, but by adopting a hybrid edge/cloud architecture, organizations can optimize performance by keeping latency-sensitive operations at the edge while leveraging the cloud for more compute-intensive workloads. This approach minimizes network congestion, improves responsiveness, and ensures data can be processed within milliseconds – whether locally or in coordination with cloud infrastructure.

This means that optimal, low latency computing is increasingly happening near the sensors and IoT devices that generate the data. For applications requiring real-time responsiveness – such as autonomous vehicles or industrial machinery – processing data closer to the source significantly enhances performance and improves safety outcomes.

Edge AI enables these benefits by handling time-sensitive tasks locally, while still allowing more complex or large-scale processing to be offloaded to the cloud as needed. This hybrid approach reduces internet bandwidth usage, lowers operational overhead, and boosts energy efficiency by limiting the amount of data that needs to travel back and forth to centralized data centres. However, sensitive information will always remain a target for cyberattacks looking to weaponize it or hold it to ransom.

Building a foundation of trust

For over twenty-five years, the Trusted Platform Module (TPM) has evolved to meet the ever-changing security needs of the computing ecosystem. Today it plays a critical role in AI and Edge Computing, where is ensures the authenticity and integrity of local data, algorithms, and operational processes. By defending critical infrastructure and endpoints against malware, firmware-level attacks, and unauthorized system changes, TPMs provide robust protection at every layer.

When deployed at the edge, TPMs provide tamper-resistant inputs, preserving system integrity from sensor to inference. This ensures protection against unauthorized modifications, safeguarding the models that power these systems and ensuring that they perform as intended, even in hostile environments. TPMs also enhance machine identity and platform integrity, crucial pillars for Zero Trust architectures and devices attestation.

Moreover, TPMs contribute to the concept of cyber resilience by enabling platforms to detect and recover from unauthorized code changes or system compromises. Features like secure authentication, protection of credentials, certificates, cryptographic keys, and user data further fortify the system. Added functionalities, such as attestation and secure boot enhance the trust of AI systems, providing a robust defense mechanism against evolving threats.

In addition, TPM-based systems support regulatory compliance through hardware-enforced security policies, reducing IT complexity and operational costs. By minimizing dependence on external security hardware like smart cards or tokens, TPMs simplify deployment and lower overheads. These advantages position TPMs as a cornerstone of modern secure computing, aligning with the broader goals of trusted and resilient technology ecosystems.

Proving our credentials

As we enter the era of Edge AI, we are seeing some valuable use cases of the TPM for protecting the data being processed. One such solution, SEC-TPM, combines hardware-based security with advanced TPM capabilities in order to protect devices and data in dynamic, complex environments, including at the edge.

Solutions like these demonstrate that the TPM is not purely just a hardware feature – they are the foundation for building secure AI applications. Organizations have the means to build trust within their edge deployments, and make them capable of withstanding any tampering or attacks, while remining compliant with the latest, stringent security standards.

Thanks to a collaboration between our members, we have been able to demonstrate SecEdge's SEC-TPM™ capabilities within an Edge AI environment. Used as a hardware Root-of-Trust, this TCG 2.0 Firmware TPM solution with activation service is currently being used to ensure the protection of AI models during both the training and development processes. TCG Compliant development kits are available for download here.

First, the application is encrypted and locked to a device in storage (such as flash memory). Once the sequence of loading and running the model is initiated, the trusted applicated hosted in a secure enclave verifies, decrypts and loads the application. By interfacing with a device’s operating system, the trusted application can then load the application directly into the Random Access Memory (RAM) and run it in a trusted state. Adopting this approach means the attack surface can be reduced, making it much safer to run an AI model.

The possibilities offered through Edge AI are vast, and it’s no surprised that the Edge AI is set to be worth approximately $356.84 billion USD by 2035. However, as more businesses turn to the concept in order to enhance operations and reduce their energy consumption, it’s vital that the TPM is used as the cornerstone for security. Of course, organizations shouldn’t just stop at the TPM – TCG offer a range of standards and specifications such as DICE and CyRes to deliver enhanced security in every facet of operations.