Your research.
Our security operations.

Security needs an owner

A GPU cluster brings together hosts, high-speed networks, shared storage, schedulers and access services. Keeping that foundation secure takes ongoing attention. We take care of security operations across the infrastructure we manage, so your team can stay focused on its work.

The industry’s recent wake-up calls

In March 2026, an AI gateway package supply-chain incident introduced credential-stealing code into two published releases. The maintainer’s report describes exposure of secrets such as cloud credentials and Kubernetes tokens. It is a reminder that a compromised dependency can reach far beyond the application that installed it. Read the maintainer’s incident report ↗

October brought separate runtime disclosures: PHP2Shell, covering two critical PHP vulnerabilities, and the Brotli decoder flaw often called HTTP2Shell or Node2Shell. These are vulnerability disclosures, not evidence that every installation was compromised. They reinforce the need to identify affected runtimes, rebuild images, deploy fixes and verify what is actually running. PHP security announcements ↗ · Node.js security release ↗

Agents make access controls even more important

An AI agent can read documents, call tools, execute code and reach external services. That makes the permissions around the model as important as the model itself. OWASP’s agent and MCP guidance highlights several risks operators need to plan for:

• Prompt injection. Instructions hidden in a webpage, document or tool response can try to redirect an agent into leaking data or taking an unintended action.
• Tool and credential abuse. An overprivileged connector or exposed token can turn a narrow task into access to unrelated systems.
• Dependency and tool poisoning. Compromised packages, tool definitions or updates can introduce malicious behaviour into a trusted workflow.
• Persistent context and unsafe actions. Poisoned memory can affect later tasks; weak approval boundaries can let agents change or delete resources without proper oversight.

OWASP AI Agent Security guidance ↗ · OWASP MCP risks ↗

How we turn that into operations

For agentic workloads, we work with your application owners to define scoped service identities, isolated execution environments, controlled network access and useful audit trails. We review the infrastructure permissions behind tools and connectors, and agree which privileged actions need independent approval. Package and image updates follow a reviewable rollout path.

These controls reduce exposure and support investigation. They complement application-level validation and agent testing; no infrastructure layer can guarantee that a model will never follow a malicious instruction.

Before an incident

We build security into the operating baseline: controlled administrative access, scoped permissions, monitoring, and a clear inventory of the systems under management. We review relevant advisories, assess exposure and plan patches around your workloads, with validation and recovery steps for changes.

When something happens

Yggdrasil provides 24/7 critical-incident response through an on-call arrangement. We take ownership of coordinating the infrastructure response, from the first investigation through containment and recovery.

• Investigate and assess. Review alerts, logs and affected systems to establish the scope and prioritise action.
• Contain the exposure. Coordinate isolation, access restrictions and credential revocation with the appropriate system owners.
• Remediate and recover. Apply relevant fixes, validate the infrastructure and return affected services to operation through an agreed recovery plan.
• Keep you informed. Communicate impact, actions and next steps, then review what happened and improve the operating baseline.

One accountable infrastructure partner

You shouldn’t have to assemble a response team while an incident is unfolding. We coordinate the work across the managed stack and involve your team or vendors when their access, decisions or systems are needed.

Your service agreement defines the managed systems, response targets, escalation channels and recovery responsibilities. That gives everyone a clear plan before an urgent situation arises.

Sources reviewed 25 October 2026. Incident details and remediation guidance can change; follow the linked maintainers for current information.