By Rekha Shenoy, CEO, BackBox
According to a recent study by Bain, 74% of companies rank AI as a top-three strategic priority, up from 60% a year earlier, and 21% say it’s their highest priority. As AI becomes more important to companies, network infrastructure vendors are bundling AI into Enterprise License Agreements (ELAs). AI can be a welcome addition to your NetOps toolbox. But how do you know if you can trust it?
Imagine your organization is building a new data center, opening an office in a new location, or acquiring another company that requires a large-scale systems upgrade. As a network infrastructure owner, you’re likely making that infrastructure investment through an ELA. ELAs often include immediate access to a broad catalog of the vendor’s devices, software, and services—whether you need all of them or not—including services that leverage AI.
Many ELAs include some functionality that goes unused. However, with Gartner predicting AI agents will be implemented in 60% of all IT operations tools by 2028, it’s only natural to ask: “What can AI do for my team?” The answer is: AI can help NetOps teams enhance security and compliance, making network infrastructure more resilient. AI can also raise concerns about reliability and the potential for misuse if you don’t proceed with caution.
Here are three requirements every vendor should meet before you trust their AI.
1. More transparency. You need to understand the data sources AI accesses and how it uses that data. Over the years, infrastructure vendors have collected millions of pieces of data in various formats. Vendors struggle to normalize that data and enrich it with context so that it maps to each customer’s unique environment and can be used reliably by AI.
According to a Harris Poll survey of more than 300 U.S. technology decision makers, nearly 90% say they can’t fully trust AI-driven insights until the data behind them is verified through formal governance. If AI is recommending actions or automatically making changes in your infrastructure, stakeholders need clear visibility into the data supporting those recommendations and actions.
2. Better explainability. Being able to question AI and get a detailed explanation for the recommendation or output is equally important. The AI model must be able to provide the sources of information and the steps it took to reach a conclusion, so you can validate the process, identify any errors, ask for alternatives, and determine the best course of action.
Explainability is particularly challenging in complex, multi-vendor environments. Suppose the AI model finds a configuration that results in a vulnerable or non-compliant situation on one device. Can the AI model from that vendor tell if your other devices from other manufacturers are experiencing the same issue or are affected? Also consider additional systems within your NetOps ecosystem that your network devices may integrate with, such as your IT service management (ITSM) and configuration management database (CMDB) tools. How they leverage AI contributes to the complexity when assessing output and making decisions, highlighting the importance of explainability.
3. Clear accountability. Ultimately, humans are accountable for outcomes, not AI. So, use cases really matter. A network infrastructure owner I spoke with explained the challenge well. The global enterprise he works for has extremely high SLAs with its customers and is not comfortable adopting AI for highly fault-intolerant use cases. Instead, they have identified use cases that are also extremely important, but where errors are less critical and can be easily corrected, or where they need to improve efficiency.
Good candidates include vulnerability risk mitigation, compliance reporting, and remediation. These use cases are a good fit because AI can help complete them more efficiently and reliably than humans alone. AI can take the mundane aspects of these tasks off your desk by quickly providing you visibility into vulnerabilities or instances of configuration drift that matter, enabling you to make better decisions.
Responsible AI use cases give you more control, not less, with humans holding accountability for decision-making. This also protects you from the dangerous habit of “cognitive surrender”, where you become too comfortable with AI, slip into autopilot, and then have to scramble to recover when an error inevitably happens. Guardrails such as designated checkpoints and exception-based alerts ensure clear accountability; humans remain in the loop to monitor progress and investigate discrepancies when needed.
Getting Value Out-of-the-Box
Adding AI tools into an ELA shouldn’t be about just deploying more software for software’s sake. It should focus on helping organizations use AI in a trustworthy and valuable way. When built responsibly—delivering more transparency, better explainability, and clear accountability—AI becomes a vital tool for NetOps teams. They can better understand what’s happening across their complex modern networks, make decisions faster, and improve network cyber resilience in ways that were previously unattainable.
