Securing AI Environments, from Data Center to Edge

In most organizations today, cybersecurity has become a top priority. As threats mount, new digital initiatives must be secure from the start of deployment. This is certainly true of artificial intelligence initiatives, and it’s particularly true of AI at the edge. Gartner has estimated that 50% of data will be processed outside the data center, and AI at the edge will be a key use case. IDC reports that over 56% of enterprises plan to deploy AI at the edge within a year.¹

AI at the edge is likely to make use of the most sensitive data an organization possesses, so a compromised AI algorithm at the edge could result in serious problems. And because “the edge” will be widely dispersed, the organization’s threat surface will increase substantially. But how can it be secured? The starting point is to use consistent edge infrastructure that can be protected with a common set of security tools and policies. And secure AI system design must utilize secure infrastructure from the outset. Trying to bolt it on after the fact is problematic.

Developing an effective security strategy for AI at the edge
The principles practiced in DevSecOps, including the recognition that security must be part of the initial planning for any system, have produced a new security mantra: “Secure by design, secure by default.” This requires secure silicon coupled with a secure and open software platform.

Another security principle that is essential to ensure secure AI at the edge is zero trust. Under this principle, no person or device, either inside or outside the organization, is trusted by default. Verification of user and device is required every time a logon is attempted. And a zero-trust approach must be paired with a defense-in-depth strategy that uses multiple layers of overlapping security tools so that, if one machine is breached, the attackers can’t gain access to your systems.

The five key elements of a defense-in-depth approach are the following:

• The use of zero-trust principles at scale across the entire distributed edge for both the onboarding and management of every device and account. And zero trust must be enforced across the entire device or account lifecycle to ensure that there are no gaps.
• The deployment of workload, model and data protection across heterogeneous nodes with trusted execution, security intents-based orchestration and harmonized application programming.
• The deployment of a trusted execution environment that balances performance and reliability. This will require crypto-acceleration to provide necessary performance within a trusted execution environment. Further, utilizing processes for identifying Common Vulnerabilities and Exposures—commonly referred to as CVEs—and leveraging the Common Vulnerability Scoring System—commonly known as CVSS—enables the team to react to newly identified threats effectively. The execution environment must also provide full telemetry data to support secure operation. Finally, implementing tools to identify and manage out-of-bound events is important as well.
• Integration of compliance/governance capabilities within the security framework. The relationship between compliance and security is very strong, and ensuring that both are part of the defense-in-depth plan is the optimal approach. This improves compliance readiness while supporting improved attestation and automated actions.
• Securing the data lifecycle at all stages, beginning with secure boot to ensure that systems boot to safe configurations and execute trusted code. Monitoring the integrity of the device, firmware and software at boot is also required. Also essential is zero-touch device onboarding to eliminate the activation of compromised or fake devices. Using AI at the edge reduces the amount of data in transit but does not eliminate it. For this reason, in-flight security protocols are still needed and important to securing data from the edge to the data center. Protecting data at runtime is possible by isolating memory with new hardware technologies and the use of role-based access controls to ensure only valid devices or users have access. The last component of securing the data lifecycle is protecting data at rest. Tamper-resistant hardware and storage, encrypted with unique hardware keys, provides greater protection.

Intel supports secure AI at the edge
Technology vendors must provide the integrated and comprehensive security tools necessary to enable their customers to better protect their critical digital resources. For example, Intel provides a combination of tightly coupled silicon and software security that constantly reinforce each other to manage the vulnerabilities that AI at the edge exposes. The use of inference at the edge enables game-changing new applications, but deploying secure and performant infrastructure at the edge is mandatory. Intel is committed to providing secure AI at the edge with numerous technologies and solutions to secure data, infrastructure and software. For more information, please click here.

¹ Infrastructure for AI at the Edge, IDC Market Perspective Dec 2022

Notices & Disclaimers
Intel technologies may require enabled hardware, software or service activation.
No product or component can be absolutely secure.
Your costs and results may vary.
Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries. Other names and brands may be claimed as the property of others.