By Guest: Alex Zenla
Company: Edera
Show Name: KubeStruck
Topic: Kubernetes
The security model we’ve relied on for containers has a fundamental flaw. Containers aren’t actually isolated environments. They’re Linux processes that simply look like containers, held together by a collection of kernel features never designed for true isolation. As AI-generated code floods production systems and GPU workloads scale, this architectural limitation is becoming a critical vulnerability.
Alex Zenla, CTO of Edera, has spent over a decade wrestling with these security challenges. Starting in IoT security at age 14 and later working at Google on industrial IoT deployments, Zenla witnessed firsthand how security becomes an afterthought when engineers prioritize functionality over protection. His solution: rebuild how we think about running containers by working below the kernel level.
From IoT Security to Cloud Native Infrastructure
Zenla’s path to founding Edera began in the Internet of Things, where the joke that the S in IoT stands for security reflects a harsh reality. Industrial IoT systems running Linux kernels five years out of date, electrical engineers deploying applications without security expertise, and devices connected to the cloud but never updated painted a troubling picture.
At Google, Zenla was brought in to solve isolation challenges for IoT systems across building deployments. The problem was clear: existing technologies like gVisor weren’t designed for the high performance requirements of industrial IoT. So she started building something new in his free time, a technology that would eventually become Edera.
The breakthrough came when Ariadne Conill, one of the original developers at Chainguard who created Chainguard Images and Wolfi, saw what Zenla was building. Her insight changed everything. The problem wasn’t just IoT. The lack of true container isolation affected the entire cloud native industry. Together with CEO Emily, they founded Edera to tackle container security at its foundation.
Building Security Below the Kernel
Edera’s approach centers on what Zenla calls a hardened runtime. While companies like Chainguard have pioneered hardened container images, Edera focuses on creating the most secure place to run those images. The distinction matters because even hardened images run on infrastructure with fundamental security limitations.
Traditional containers are essentially a collection of Linux kernel features, namespaces, cgroups, and other mechanisms cobbled together to create the appearance of isolation. The kernel itself doesn’t understand containers as a distinct concept. This creates security boundaries that can be breached because applications are still just processes sharing kernel space.
Edera’s technology works differently. By operating below the kernel level and introducing a new concept called zones, the platform provides true hardware-level isolation. This architectural shift allows Edera to guarantee that workloads running on the same hardware cannot interfere with each other, even if one is compromised.
The results from a recent Trail of Bits security audit validate this approach. Edera showed no medium or high vulnerabilities, and the audit calculated that the platform reduces code running in ring zero, the lowest and most privileged layer of hardware, by 95 to 97 percent. Reducing the attack surface at this level represents a fundamental improvement in container security.
Rethinking GPU Security for AI Workloads
The rise of AI workloads has exposed another critical gap in cloud native security. GPUs don’t work like CPUs, and the industry has largely accepted that Nvidia’s massive proprietary drivers will exist as potential security weaknesses in production systems. Zenla refuses to accept this compromise.
Edera is building virtualization capabilities specifically for GPUs, creating the same container-like isolation mechanisms but for graphics processors. This approach addresses multiple challenges simultaneously. For training and inference workloads, GPU pass-through provides direct access while maintaining isolation. For multi-tenant GPU clouds, virtualized GPUs enable splitting resources without sacrificing security.
The performance implications are significant. Many GPU deployments suffer from poor utilization, with expensive hardware sitting idle. Edera’s GPU virtualization allows cloud providers to maximize resource usage through secure multi-tenancy while maintaining the performance characteristics AI workloads demand.
Beyond GPU isolation, Edera addresses the emerging challenge of AI-generated code. As development teams deploy code written by large language models without thorough human review, the risk of introducing vulnerabilities multiplies. Running AI-generated applications in truly isolated environments means a security flaw in one component can’t cascade across the infrastructure.
Recent research by Marina Moore, an Edera researcher and member of CNCF’s TAG Security, demonstrated that AI agents running on Edera actually perform faster than on containerd. The performance gains come from low-level hardware access and micro-optimizations in CPU and memory scheduling that are only possible when working directly with hardware.
Practical Innovations: From Boot Times to Kubernetes Integration
Edera’s work extends beyond runtime security into fundamental system improvements. The team recently released Sprout, a new Linux bootloader designed to replace GRUB, which has been the standard for 20 years. On complex hardware, GRUB can take minutes to boot. Sprout reduces this to milliseconds, a critical improvement for AI workloads and edge deployments where rapid scaling matters.
The platform also integrates directly with the cloud native ecosystem. At KubeCon Atlanta, Edera announced integration with Falco, allowing system call events to stream directly from the Edera hypervisor into Falco engines for real-time threat detection. This demonstrates how working at the hypervisor level enables security capabilities impossible with traditional container runtimes.
For Kubernetes deployments, Edera’s technology addresses the platform’s preference for larger nodes while still supporting resource-constrained environments. The low-level hardware access enables better bin packing and resource utilization. For edge devices too small for Kubernetes, Edera can deploy single applications with hardware-level protection, making the security model flexible across deployment sizes.
Contributing Back to Open Source
Despite building proprietary technology, Edera maintains deep connections to the open source community. The team contributes actively to Xen and the Linux kernel. Zenla herself has discovered and patched CVEs in the Linux kernel. The company packages Linux kernels as OCI images, leveraging container standards for system deployment and management.
This commitment to open source collaboration extends to working with Linux Foundation projects around security and isolation, though Edera isn’t trying to compete with existing initiatives. Instead, the company positions its technology as complementary, providing a more secure foundation for the tools and platforms the community already uses.
The philosophy reflects Zenla’s 12 years in open source development. Solutions that work in isolation don’t create lasting change. By contributing improvements like Sprout and working within established ecosystems like Kubernetes and Falco, Edera can improve security outcomes across the industry rather than just for its own customers.
