As Generative AI (GenAI) and Agentic AI rapidly permeate enterprise ecosystems, they are reshaping not just innovation and productivity but also the cybersecurity landscape. While GenAI tools like ChatGPT and Google Gemini focus on creating content based on prompts, Agentic AI goes a step further, making autonomous decisions, executing tasks, and learning from outcomes like a virtual co-worker. With these capabilities come new and complex security risks ranging from inadvertent data leaks to unchecked access privileges and malicious model manipulation. In this exclusive conversation with Tech Achieve Media, Ajay Gupta, Vice President and Country Manager – SAARC at Netskope, shares critical insights on how organizations can adapt their security postures in this AI-powered era, implement robust Zero Trust strategies, and ensure data protection without compromising performance or agility.
TAM: With the rapid adoption of Generative AI and Agentic AI in enterprises, what are the most pressing security vulnerabilities you foresee?
Ajay Gupta: If we try to understand and differentiate between Gen AI and Agentic AI, here’s how it works:
Gen AI is often described as reactive. You provide a prompt, and it returns a result, which could be new text, images, source code, etc. However, for the next step, the user typically needs to intervene to move things forward.
Agentic AI, on the other hand, can perform actions. You give it a prompt, and the result can be executable or actionable. Agentic AI doesn’t just respond but can decide what actions to take based on the prompt, execute those actions, learn from the results, and continuously update itself. In that sense, it acts like a co-worker capable of completing entire tasks within an organization.
That’s a significant shift. Within enterprises, many have started exploring Gen AI capabilities and are developing multiple solutions using it. In light of Gen AI usage increasing among organizations, Netskope now has more than 1550 top Gen AI applications similar to platforms like Google Gemini or ChatGPT in its Cloud Confidence Index, allowing organizations to understand each application’s risk profile, and define security policies accordingly. These engines offer a range of capabilities.
However, organizations face major challenges, especially with data leakage in Gen AI. Often, employees unknowingly upload critical data like PII or source code to these platforms. Developers, for example, might paste source code to validate or debug it. Others may enter confidential information without realizing the implications.
This creates the need to clearly differentiate between corporate-sanctioned Gen AI tools and personal ones. For instance, organizations may allow the corporate version of ChatGPT but restrict personal accounts. Netskope plays a vital role here, which is to help organizations monitor, manage, and enforce policies on sanctioned Gen AI apps. It defines what information can be safely shared with tools like ChatGPT and helps mitigate risks.
When it comes to Agentic AI, the challenge shifts from data leakage to data access control. Since Agentic AI behaves like a co-worker and can take autonomous decisions, it’s critical to define what roles it plays and what data it can access. In fact, CIOs or CISOs must now take on a role similar to that of an HR leader ensuring that the AI “employee” is assigned clear responsibilities, monitored for performance, and only granted appropriate data access.
Giving too much access can lead to data exfiltration, just as with human employees. CIOs must ensure Agentic AI tools only access what’s necessary to perform their tasks and nothing more.
To summarize:
- Gen AI’s key challenge: Preventing data exfiltration.
- Agentic AI’s key challenge: Managing data access appropriately.
These are the core issues organizations are dealing with today, and Netskope is helping them address these challenges effectively through its platform.
TAM: How can organizations mitigate risks such as model poisoning and adversarial attacks?
Ajay Gupta: We are doing a lot of development in this area, especially around model poisoning and other types of attacks. From an organizational perspective, access control is crucial.
As mentioned earlier, data pipelines should not be accessible to the external world. They must operate in a controlled environment. Any breach, whether it’s malware or another form of attack, can lead to unintended and potentially dangerous actions being taken by Agentic AI. This is why strict access control mechanisms are essential.
The second important aspect is regular audits of data pipelines. Netskope provides SaaS Security Posture Management (SSPM), which is critical because these pipelines often run in cloud environments. SSPM ensures the SaaS applications are properly configured. It checks for the correct implementation of rules around sensitive data like PII, PCI, source code, and others. It also validates whether password policies and compliance standards such as NIST are being followed. All these compliance elements come into play. The Netskope platform supports regular audits to identify and fix gaps. This helps strengthen the overall security posture and ensures the environment remains protected.
TAM: Agentic AI, characterized by its ability to perform autonomous tasks, raises new security concerns. How can cybersecurity frameworks evolve to detect and counter unauthorized actions by such systems?
Ajay Gupta: With Agentic AI, as you mentioned, it can take a lot of actions. The main concern is access control.
Netskope’s Zero Trust Network Access helps define what kind of access Agentic AI can have and what actions it can perform. You cannot give it an entirely open environment, as unrestricted actions can lead to critical problems for the organization.
Rules need to be clearly defined to limit the actions Agentic AI can take. Beyond those limits, it should not be able to execute commands. Netskope’s Zero Trust Network Access ensures control over which applications can be accessed and what actions can be executed after learning from and processing commands.
The AI operates within a restricted profile. It should not be granted unrestricted capabilities. Similarly, for a CIO, it’s like onboarding a new employee. When someone joins an organization, their access is defined based on their role. You determine which applications they can access, what data they can upload, and what responsibilities they have.
The same principle applies to Agentic AI. The CIO must define its access boundaries to ensure it performs the intended tasks securely. The management team should also be informed about which databases the AI will access and the expected output from its role.
This approach aligns with a Zero Trust Network Architecture, specifically designed to control and limit the actions of Agentic AI. It helps minimize unnecessary data access, reduces risk from threat actors, and prevents misuse such as the installation of malware. By using Zero Trust architecture, the AI’s profile is capped and kept within safe operational boundaries.
TAM: As organizations deploy AI-driven cybersecurity solutions, how can they ensure transparency, build trust among stakeholders?
Ajay Gupta: To be very frank, in today’s world, any security tool must have AI and ML engines built into it. If you’re not building that, you’re becoming obsolete. Any modern security solution today needs to have AI and ML running behind it. Your question is more from a security vendor’s perspective. Yes, they have to build it. Because there is too much data to handle manually. One of the biggest challenges organizations face today is the overwhelming number of alerts. Logs are huge, and the response time from internal teams, like the security operations team, can be slow. With AI and ML engines, these tools can prioritize which incidents need immediate attention and guide the right people on what actions to take.
I haven’t seen any organization ignoring this need. But one of the most important aspects is transparency. Security vendors must be clear with organizations and customers about what tools are being used and what data is being processed to generate results. For example, at Netskope, we have a dedicated team of more than a hundred engineers and data scientists working on AI initiatives, with over 40 patents in AI and machine-learning. It’s a very experienced team. We also have a published URL that details the data we use and the methodologies applied. This level of transparency builds trust. Organizations need to understand what data is used and how decisions are made. It’s critical for AI adoption in security.
Transparency and explainability are key. We provide public documentation that explains how decisions and actions are derived. This is essential so that organizations know what to expect and can build confidence in the outcomes. At Netskope, we have invested heavily in this area. We hold over 275 patents, with more than 40 of them specifically related to AI, developed by our in-house teams. A significant part of our R&D investment is focused on advancing AI and ML capabilities in cybersecurity.
TAM: AI is often viewed as a double-edged sword in cybersecurity. How is Netskope leveraging AI to proactively identify and counter sophisticated threats, and what innovations are you pioneering in this space?
Ajay Gupta: Within organisations, what we’ve observed over the last year, based on findings from our threat labs, is a 30x increase in the amount of data being uploaded by employees to AI applications when they are at work. That’s a massive jump.
With the Netskope platform, built on Zero Trust architecture, we help secure this environment. New SaaS applications continue to emerge at a fast pace, and Netskope evaluates more than 83,000 of them using a Cloud Confidence Index, which scores each application based on several security perimeters.
For example, if an application receives a score below 50, it’s flagged as risky. Some applications score as high as 90 or 95 and are considered safe. These scores are determined based on factors defined by the SaaS provider, such as handling of PII, compliance readiness, access control, and other risk indicators. Netskope uses AI engines to support this scoring and classification process.
What we’ve seen within organizations is that IT typically sanctions only 5 to 10 applications, but in reality, employees use over 1,000 on average, many of which IT is unaware of. The Netskope platform provides full visibility into which applications employees are actually using. Based on the risk score, IT can choose to allow, block, or educate users accordingly.
Education is a key part of this. Often, employees don’t intend to do anything wrong, they just don’t know the company policies. Netskope helps by identifying risky or unsanctioned apps and guiding employees to use approved alternatives, enabling them to meet their business needs securely.
With Netskope One, organizations can reduce the risk of using unsanctioned or risky applications and enforce Zero Trust-based access. Even if a device is compromised, a threat actor will not be able to access critical applications, thanks to tight access control.
The biggest advantage of Netskope is its unified approach:
- Single agent
- Single user interface
- Single policy framework
This significantly reduces the number of agents needed across the organization and simplifies operations. For example, a single policy can cover DLP, SaaS, IaaS, email, and more, reducing both complexity and operational costs.
Another major benefit is that skilled cybersecurity talent is hard to find. Netskope simplifies the security infrastructure so that organizations can achieve robust protection without needing large, specialized teams. This is the feedback we consistently hear from our customers.
Netskope helps organizations reduce the number of agents while improving their security posture without compromising on performance or user experience. That’s critical.
A few years ago, security tools often degraded performance. Users would complain that security measures slowed them down. But Netskope has built 117 data centers across 75 regions, ensuring low-latency performance and a seamless user experience.
Even for roaming users, there is no need to connect to a VPN and route traffic back to the corporate data center. They can connect directly to the internet through Netskope’s global infrastructure with all security policies still enforced. This combination of performance and security is what Netskope delivers. While network teams focus on performance and CISO teams focus on protection, Netskope bridges both priorities without compromise.
TAM: Looking ahead, how do you envision the role of AI in shaping the future of cybersecurity? Are there any emerging trends, such as federated learning or explainable AI, that you believe will redefine the industry
Ajay Gupta: When you look at the Secure Access Service Edge (SASE) market and third-party reports like those from Gartner, you’ll notice a significant influx of players entering this space. This is because every industry is moving toward consolidation and adopting the SASE framework as part of their digital transformation.
If you refer to the Gartner Magic Quadrant for SASE, Netskope has consistently been positioned in the Leaders quadrant since the report’s inception. That leadership is driven by two key areas where we are heavily investing: Artificial Intelligence/Machine Learning (AI/ML) capabilities and data protection.
We have over over 275 patents, with more than 40 specifically focused on AI. This demonstrates our commitment to building intelligent engines that power advanced threat protection, automation, and data security. The second critical area is data. Whether it’s traditional AI or Generative AI (GenAI), the primary concern is that data should not be exploited or mishandled. With a Zero Trust Architecture (ZTA), we ensure users do not have over-privileged access, reducing the risk of misuse or unauthorized actions.
Netskope is investing heavily in:
- Security Service Edge (SSE) as a core component of our SASE vision.
- Simplifying cloud transformation for organizations. Many businesses view cloud migration as complex, but it doesn’t have to be. Netskope enables secure, smooth transitions to the cloud.
- Securing data across all touchpoints. In today’s environment, data is everywhere. Users are accessing it from managed devices, unmanaged devices, mobile phones, and more. Access rights must vary depending on the device and user profile, especially when using corporate vs personal machines.
We empower organizations to securely enable user access while keeping data exfiltration risks in check.
Another critical area we focus on is user experience. Security should never come at the cost of performance. Users should be able to access data and applications, whether Software-as-a-Service (SaaS), on-premises, or cloud-hosted infrastructure-as-a-service (IaaS), from any device, anywhere, with high performance and strong security.
Our platform ensures visibility across the entire path so organizations can pinpoint exactly where latency occurs. This enables a smooth, reliable experience for end users, which is a top priority. Netskope also stands out in Data Loss Prevention (DLP). According to third-party reports, our comprehensive DLP offering is a major differentiator. We cover five key vectors:
- Web
- Endpoint
- IaaS (Amazon Web Services [AWS], Google Cloud Platform [GCP], Microsoft Azure)
- On-premises and SaaS applications
Across all these channels, we run advanced AI/ML engines to help organizations protect and secure their data effectively. This combination of advanced technology, holistic coverage, and seamless performance is what gives Netskope a leading edge in the market.
