Hopefully, they can deliver because AI programs like Anthropic’s Mythos AI are cracking open programs faster than an otter can shuck oysters.

MINNEAPOLIS — At Open Source Summit North America, Edera, a leader in hardened runtime isolation for containers, GPUs, and AI agents, and Minimus, a top company in hardened container images and software supply chain security, announced they were teaming up to deliver what they claim is true end‑to‑end container security for critical infrastructure operators. The plan is that by combining hardened container images with a hypervisor‑backed runtime designed to stop attacks even when vulnerabilities slip through, they provide you with the software security you need in an ever more dangerous AI world. 

How bad is it? In a statement, Ben Bernstein, CEO and co-founder, Minimus, said “AI-powered vulnerability discovery has changed the math on open source risk. The question isn’t whether adversaries will find exploitable flaws in widely deployed software – it’s how fast, and what happens next.” Their answer: “Minimus exists to eliminate and shrink the target as aggressively as possible. Edera exists to ensure that what can’t be shrunk away can’t be weaponized. This partnership puts both layers in front of the organizations that can least afford a breach to become an outage.”

Specifically, the partnership brings together Minimus’s minimal, continuously patched container images and Edera’s “hardened runtime,” which runs each workload inside its own micro‑VM. The latter eliminates the shared‑kernel model that has long worried security teams in regulated sectors. The pair believe financial services, federal agencies and operators of critical infrastructure that face strict compliance regimes and a rising tempo of AI‑assisted attacks will find their approach just what they need. 

Minimus, whose founders trace their roots to early container security efforts, focuses on shrinking the attack surface at build time. Its platform rebuilds images from upstream source, strips out non‑essential components and maintains what it describes as a near‑zero‑CVE posture through continuous patching and signed software bills of materials (SBOMs). The goal is to ensure that by the time a container is deployed, it carries as little inherited risk as possible.

Edera attacks the other side of the problem: runtime isolation. Rather than relying on namespaces and cgroups atop a shared Linux kernel, Edera interposes a thin hypervisor layer and runs each container or AI workload inside a dedicated micro‑VM. That approach, similar in spirit to lightweight virtual machines, is designed to prevent container escapes, kernel‑level privilege escalation and lateral movement between workloads sharing the same node.

By integrating the two stacks, the companies argue they can offer a more credible defense‑in‑depth story than traditional scanning‑centric pipelines. Minimus images aim to eliminate known vulnerabilities and reduce bloat before deployment; Edera then assumes something will eventually be exploited anyway and focuses on containing any compromise within a strictly bounded runtime environment. Together, the combination is being positioned as a way to reduce both the probability of compromise and the blast radius when it happens.

The partnership also leans into the boom in AI and GPU‑accelerated workloads. Edera has already been marketing its hardened runtime for GPU and agentic AI scenarios, where direct hardware access and complex dependency chains can amplify risk. Pairing that runtime with Minimus’s slimmed‑down, source‑rebuilt images is meant to appeal to organizations experimenting with AI in sensitive environments but wary of the security implications of rapidly assembled container stacks.

Both vendors are aligning their messaging with established guidance, such as NIST’s container security recommendations and federal cloud requirements, in hopes of tapping into modernization and zero‑trust budgets across government and critical infrastructure. 

While details on joint go‑to‑market motions and pricing were not disclosed, the move underscores a broader trend among buyers toward tightly integrated combinations that address the software supply chain and runtime behavior as a single solvable security problem. Will this pairing work? With the tempo of AI-based hacking attempts ever increasing, we’ll soon find out.