The Speed Asymmetry That’s Breaking Enterprise Defense

Cybersecurity has entered a new era of crisis: vulnerabilities are being discovered and exploited faster than organizations can patch them. Mandiant’s M-Trends 2026 report reveals a sobering statistic—28.3% of CVEs are now exploited within 24 hours of disclosure. This represents a fundamental shift in the threat landscape, where the traditional assumption that patches arrive before exploitation no longer holds.

The timing couldn’t be worse for European enterprises scaling AI infrastructure. As organizations rush to deploy large language models and AI agents, they’re inheriting a double burden: the discovery rate of vulnerabilities is accelerating, but patch deployment tools and processes remain underdeveloped.

What’s Driving This Vulnerability Explosion

The recent disclosure of critical vulnerabilities in Microsoft’s Semantic Kernel—affecting over 27,000 projects on GitHub—illustrates the problem in real time. Semantic Kernel SDK versions prior to 1.71.0 contain prompt injection flaws that could allow remote code execution. Yet many organizations won’t know they’re vulnerable until well after exploitation begins.

This isn’t a Microsoft problem. It’s systemic. The volume of vulnerability discoveries has increased dramatically across the entire software ecosystem, but enterprises have not deployed corresponding infrastructure to track, prioritize, and patch these issues at scale.

Why Irish and European Enterprises Face Unique Risk

For organizations operating under the EU AI Act’s August 2026 deadline, this vulnerability lag presents an urgent compliance challenge. High-risk AI systems must demonstrate robust security controls by that date. But if vulnerability discovery and exploitation are now outpacing patch deployment, how can enterprises credibly claim their AI infrastructure is secure?

Irish tech firms, which host significant AI infrastructure in European data centers, face particular pressure. Many are self-hosting LLM frameworks to reduce latency and maintain data sovereignty. Yet the research shows that self-hosted infrastructure is being deployed with minimal authentication, no default security hardening, and unclear patch management workflows.

The Practical Problem: Detection Without Remediation

Here’s where the market is failing: enterprises have visibility tools that detect vulnerabilities but lack the orchestration layers to fix them at scale. Security teams can identify that their Semantic Kernel deployment is vulnerable, but integrating that finding into their deployment pipeline, testing the patch, and rolling it out across distributed services takes weeks—if not months.

Meanwhile, exploit code for disclosed vulnerabilities appears within hours.

What Organizations Should Do Now

  1. Audit your AI framework inventory: Identify all instances of Semantic Kernel, LangChain, and other frameworks running in production. Know your versions.

  2. Implement vulnerability detection in your CI/CD pipeline: Don’t wait for security scans. Catch vulnerable dependencies before they deploy.

  3. Create an emergency patching protocol: For critical CVEs affecting AI infrastructure, establish a fast-track review and deployment process.

  4. Prioritize by risk surface: AI agents that execute code (like Semantic Kernel agents) should be prioritized for immediate patching.

The Larger Question

This isn’t just about keeping up with patches. It’s about whether enterprises can realistically secure AI infrastructure at the pace it’s being deployed. The 24-hour exploitation window suggests they can’t—not with current tools and processes. That’s a problem the EU AI Act compliance timeline doesn’t yet address.

Source: Mandiant M-Trends 2026 Report