Employees are connecting enterprise data to unauthorized AI tools every day. AI gateways intercept some traffic, but models running on personal devices, unofficial endpoints, and alternative channels fall outside their reach. No vendor has fully solved this yet, and the attack surface expands every time a new AI tool reaches the consumer market.
In this interview on TFiR, Rajesh Ganesan, CEO at ManageEngine, breaks down why shadow AI is the defining unsolved problem in enterprise security today, what a genuine AI control plane must cover, and why bolting AI onto existing products compounds the risk instead of reducing it.
Guest: Rajesh Ganesan, CEO at ManageEngine
Show: The Agentic Enterprise
LinkedIn: Rajesh Ganesan
Watch the discussion on YouTube
Here is what every security architect and IT leader needs to know.
Technical Deep Dive
Q: Why is shadow AI now a critical enterprise security problem?
Rajesh Ganesan, CEO at ManageEngine, explains that employees under constant delivery pressure will find ways to use AI tools regardless of corporate policy. Because AI is so compelling as a productivity tool, workers actively look for channels that circumvent approved workflows. The result is data flowing to AI systems, including models running locally on personal devices, that IT has no visibility into and no control over.
“AI is so compelling. We are all constantly under pressure to deliver. Companies have policies and procedures governing how AI should be used, but employees are always looking for avenues and opportunities to circumvent them and find alternative ways of doing things.”
Rajesh Ganesan, CEO, ManageEngine
Q: What does an AI gateway actually protect against, and where does it fall short?
Ganesan acknowledges that ManageEngine and others have built AI gateway capabilities that can intercept traffic and apply controls. The gap is endpoint diversity: a language model can run directly on a personal phone, completely bypassing any network-level gateway. Covering every channel, every device type, and every AI endpoint is a problem no vendor has fully solved as of this interview.
“We can build an AI gateway and route all traffic through it, allowing us to intercept that traffic and perform certain actions. But people are smart, right? A model could be running directly on someone’s phone. How do you control that, especially if the device could be either an unauthorized phone or an approved one?”
Rajesh Ganesan, CEO, ManageEngine
Q: What must a comprehensive AI control plane cover to be effective?
Ganesan frames the requirement as holistic coverage: the control plane cannot protect only one channel or one approved endpoint. It must look into every part of the infrastructure where data flows and track which AI systems that data reaches. ManageEngine has made progress toward this architecture but Ganesan is explicit that the full solution is still in progress industry-wide.
“It has to be holistic. You cannot protect just one channel or avenue. You have to look into every nook and cranny of your infrastructure, understand where data flows, and identify which AI systems it is being sent to.”
Rajesh Ganesan, CEO, ManageEngine
Q: Why does bolting AI onto existing products create additional enterprise risk?
When AI is added as an afterthought, it inherits none of the design decisions that make enterprise software trustworthy around data sovereignty, compliance, and cost predictability. Ganesan draws a direct parallel to security: the same mistake enterprises made by treating security as a late-stage feature is being repeated with AI. Bolting AI on introduces token economics, data protection exposure, and sovereignty questions that the underlying product was never designed to handle.
“When you simply introduce AI, it may deliver certain benefits, but it will also bring additional challenges and headaches, including issues around sovereignty, data protection, and token economics.”
Rajesh Ganesan, CEO, ManageEngine
Q: How does ManageEngine approach AI integration differently?
ManageEngine starts from the customer’s desired business outcome, not from AI as a capability to add. Ganesan describes the company as a non-FOMO company: not racing to ship AI features, but working to understand what problem the business actually needs solved and what economic model fits. ManageEngine builds its own language models and has chosen not to price AI as a separate line item, treating it instead as a productivity feature embedded in the product.
“We are not in any race, including the AI race. We are simply going to use AI as a powerful tool to solve the problems faced by modern-day enterprises.”
Rajesh Ganesan, CEO, ManageEngine
Q: Is AI security now part of enterprise cybersecurity strategy, or still separate?
Ganesan places AI security squarely inside the broader cybersecurity problem set. He states that technology leaders are thinking about how to secure their environments for AI before they deploy AI, not after. He frames the AI control plane challenge as the next phase of enterprise security, comparable in scope and difficulty to the shift that made security a first-class engineering concern.
“Even before deploying AI, technology leaders are thinking about how to secure their environments to better leverage AI.”
Rajesh Ganesan, CEO, ManageEngine
Resources & Documentation
- ManageEngine, enterprise IT management and security platform from Zoho Corporation
***
👇 Click to Read Full Raw Transcript
Swapnil Bhartiya: Although security has now changed. It used to be an afterthought, it used to be someone else’s problem. Now security. You have to start thinking of security from the day one when you’re writing the piece of code. Because the way we are deploying, it’s too late by that. It’s like thinking about putting brakes or airbags after you sold the car. It doesn’t make, that model doesn’t work. Now when it comes to security or consolidation, that is more or less like evolution of the product journey. AI is something that has been just thrown at us all. So how are you or manage engine thinking about either embedding AI into your products or like bolting on or building it as a standalone capability. How do you
Rajesh Ganesan: that approach fundamentally is something we don’t want to take. Throw a into something or bolt a into something. Then we are basically sort of how to say like you said, security cannot be an afterthought. Similarly, AI cannot just be bolt in into the things. So this approach is something we are not going to go after. We want to start with understanding what our customers really want. You have a as a challenge, you have security as a challenge, but they are all just challenges. But what does the business really want? What is their outcome? What is the process that they have? What is the kind of model they want to go with? So what is their economics to doing all this? We start from there. So like how I said, we evolved into doing cyber security the right way, which is not as an afterthought. We are applying the same principle to doing AI also. In that sense we are more like a non FOMO company. I learned this word just now. We are not in any race, we are not in any AI race. We are just going to use AI as a very powerful tool to solve problems of modern day enterprises. So that is the approach that we take to that count. When you just throw in AI, it may help you with certain benefits, but it is also going to bring you additional challenges, additional headaches. Things like sovereignty, things like data protection, things like token economics. How does a company handle all that when you just throw AI in? So we want to consider all these aspects. We build our own language models. We have not praised our AI yet. So we are thinking to never praise it because it just additional feature inside your product which helps you with better productivity. Why would you look at AI separately, make it charge separately, all of that. So we take that approach. We want to bring that approach to how we do AI the way it benefits our customers. So that is the path we are taking.
Swapnil Bhartiya: in companies, employees bring their own AI tools into the workflow. How are your customers through manage engine managing and securing that layer? Because we are not even talking about regulatory industries because you have to be but a small slip. And also when you’re using API, all the data, if you’re not using, even you use zdr, nobody knows what the company is actually doing. So how do you ensure that when you give them this flexibility, the layer is secure?
Rajesh Ganesan: Yeah. So I’ll be honest, this is a problem that we are still working on. I don’t think anyone has cracked it yet. You talk about people bringing their own AI or as we call it the shadow AI as we speak today. We do have some controls. We can build an AI gateway and make all traffic go through that gateway, intercept and do certain things. But people are smart, right? And what kind of endpoints they use and what channels they use. Because AI is so compelling. We are all always under pressure to deliver things. The companies would have policies and procedures on how to use AI. The workforce, the employees are always looking at avenues, opportunities to sort of see circumvent other ways of doing things. Also this is one of the hardest problems. When I said cyber security number two, it includes doing AI security properly. So even before deploying AI technology leaders are thinking about how do we secure our environment to leverage AI better. So to that count, what we are thinking inside managed engine, how do we build a very comprehensive, capable a control plane. Right. So it has to be holistic. It cannot protect only one channel or one avenue. It has to look into every nook and corner of your infrastructure where data flows, which AI systems it goes to. Because a model could be sitting inside your phone, right. How do you control one of it could be from an unauthorized phone, an approved phone. These gaps would exist. This is what we need to address. So we have started working, we have made a lot of progress. So that is where we are headed. In terms of your question of how do companies protect their workforce from all the capabilities that AI provides, all the channels that AI provide. It’s not an easy problem. This is going to be the next, I would say the typical phase, the next million dollar problem in security would be doing this right?





