Why Securing Operational Access Is Critical to AI Data Center Resilience

The Overlooked Foundation of AI Data Center Security

Data centers are critical infrastructure, and cybersecurity is essential to their resilience and operational continuity. This is the central argument of an important recent article by Leo Simonovich of Siemens Energy and Filipe Beato of the World Economic Forum. 

For years, conversations around data center security focused primarily on protecting IT systems, such as servers, applications, networks, and cloud environments. But today’s AI factories and data centers depend on unprecedented levels of power, cooling, and operational infrastructure. This rapidly growing reliance on operational technology (OT) is creating new attack surfaces where cybersecurity failures can directly impactinfrastructure stability and uptime as well as the many industries and services that now rely on AI. 

So what does it take to avoid such failures? 

Simonovich and Beato argue that data center resilience is closely linked to energy resilience and emphasize the need to safeguard power systems and related infrastructure. They write, “A data centre’s power plant or backup battery systems should be protected – or ‘hardened’ – through correctly configured digital controls.” 

This is certainly true, but it raises a crucial question: what do those digital controls look like in practice? 

Hardening systems, improving monitoring, and expanding visibility into interconnected infrastructure, as the article recommends, are all necessary measures. However, they address only part of the challenge. 

At its foundation, long-term data center resilience requires controlling access to the operational systems that power AI infrastructure. 

AI Data Centers Are Industrial Environments 

Behind every AI workload sits a complex ecosystem of both legacy and more modern operational technology, including electrical distribution systems, generators, cooling infrastructure, water systems, building management systems, fire suppression controls, and more.  

If any one of these systems fails — whether due to equipment malfunction, operator error, or cybersecurity incident — the consequences can be immediate and severe. For instance, cooling instability can force workloads offline, while disruptions involving power or environmental controls can jeopardize operations and service availability.  

Ensuring data center resilience therefore means securing the full operational ecosystem.  

Resilience in AI Data Centers Starts with Access Control 

Operational access is often the entry point into critical OT systems inside the data center. In these highly interconnected environments, a single unmanaged vendor session or overly broad access pathway can create exposure across power, cooling, and other operational systems. In practice, compromising the infrastructure supporting AI workloads may prove far easier than targeting the workloads themselves, while ultimately causingthe same disruption to uptime and service availability. 

This is why AI data center cybersecurity must be built around strong controls over who can access operational systems and what they can do while connected. Such controls are especially important in mid-tier colocation facilities, enterprise data centers, and unmanned environments, where third-party and remote access are indispensable to daily operations. 

And access control is no longer just a security and operational concern — it is increasingly becoming a compliance requirement as well. Across critical infrastructure sectors, regulations and industry guidance are placing greater emphasis on secure remote access, third-party vendor access, and tighter controls around OT systems and assets. 

In Europe, mandates such as NIS2 now formally classify many data centers as essential infrastructure operators, heightening expectations around OT governance, remote access security, vendor oversight, and operational resilience. As AI data centers become even more intertwined with energy infrastructure and grid resilience, regulatory demands will only intensify. 

The Third-Party Access Challenge Inside AI Data Centers 

Modern AI data centers depend heavily on third-party vendors for maintenance as well as ongoing support of critical operational systems, often through both remote and on-site access. 

In many data centers, including enterprise and colocation facilities, third-party specialists may be the only personnel with expertise in specific power, cooling, or building management systems. These facilities tend to rely on continuous third-party involvement to optimize performance, monitor infrastructure health, troubleshoot system interactions, and support live workloads in real time. In such environments, vendors are essential to day-to-day operations and frequently maintain persistent or recurring access to sensitive operational infrastructure. 

Hyperscale data centers — the proprietary facilities often operated by major technology providers — typically require less third-party involvement. But even in these more mature operations, vendors still need periodic access to critical systems for maintenance and specialized services. 

This operational dependence on external personnel creates a unique security challenge. To keep data center operations running, third-party vendors need privileged access to systems that directly affect uptime and infrastructure stability. But unlike internal staff, vendors are not always bound by the organization’s corporate security policies and are more likely to use unmanaged or unmonitored devices. Given the broad and often minimally supervised access vendors maintain, a single mistake or unauthorized action could quickly affect systems across the facility and beyond. 

What Secure Access to AI Data Centers Should Look Like 

The key to implementing strong, practical access controls is to ensure they do not create friction or slow operations. Users should receive only the permissions needed for their specific roles, allowing them to complete their tasks efficiently without accessing unrelated systems. 

Effective operational access control is built on several core principles: 

1. Identity-based, zero-trust access. Every connection to OT systems should be tied to a verified identity, not a shared account or broad network pathway. Access should also be limited by role, system, and time. 

2. Real-time visibility and session monitoring. Security and operations teams need full visibility into who is connected to critical assets and what actions they are performing. For high-risk connections, including third-party access, supervisors should be able to intervene before a mistake or misuse turns into an outage.  

3. Operational data sovereignty. Operators should not have to expose operational data, infrastructure telemetry, credentials, or access activity in order to enable secure remote access. Zero-trust access models that keep all data within the customer’s environment help reduce both cybersecurity exposure and compliance risk. This is becoming increasingly important as regulatory scrutiny around critical infrastructure and energy resilience continues to grow. 

4. Segmentation. Operational environments should be segmented so that an incident impacting one system does not create unnecessary exposure across the rest of the facility. To maximize resilience, power systems, cooling infrastructure, building management platforms, and other OT systems should not share unrestricted connectivity or broad access pathways. 

5. Consistent governance across remote and on-site access. Access controls must be applied across both remote and local operations. An employee connecting remotely and a contractor working inside the facility can each introduce risk if access is overly broad or poorly controlled. 

AI data center resilience relies on continuous and secure access to operational systems. This means adopting tightly governed access controls, real-time oversight of operational activity, and clear segmentation between critical OT systems and networks. If these measures are not in place, the operational infrastructure supporting AI workloads can quickly become a pathway to disruption and instability. 

Simonovich and Beato are right to assert that cybersecurity is now fundamental for data centers. But resilience requires more than hardening systems and improving visibility. As AI infrastructure increasingly takes on the characteristics of critical infrastructure, operational access control must form the foundation of data center cybersecurity.