The rapid arrival of agentic artificial intelligence is forcing cybersecurity teams to rethink how security operations centres (SOCs) detect threats, investigate incidents and make decisions. While vendors increasingly promote AI agents capable of automating complex security tasks, questions remain over how effectively enterprises can integrate these capabilities without creating new operational risks.
The challenge extends beyond selecting the right technology. Organisations need to consider data sovereignty, model reliability, operational continuity and the appropriate division of responsibility between humans and autonomous systems. For smaller security teams and managed security service providers, AI could also fundamentally change skills requirements and established operating models.
To explore what this transition means in practice, Digital Journal spoke with cybersecurity expert Monzy Merza (co-founder and CEO of Crogl) about the realities behind agentic AI, the dangers of taking vendor claims at face value, and why successful adoption may depend less on sophisticated technology than on people, processes and clearly defined security outcomes.
Digital Journal: How do you expect agentic AI to change security operations centers (SOCs) in the near term, and should organizations take vendors’ promised capabilities at face value?
Monzy Merza: There is tremendous opportunity with AI in the SOC. But in the near term, agentic AI will create more work, more confusion, and general dissatisfaction. Enterprise security teams are falling into the same trap as their big data and cloud predecessors. They are skipping the crucial step of defining the outcomes, and that’s just the tip of the iceberg.
What’s below the surface is even more important. Security teams and enterprise governing boards are still learning the basics of AI. They are unfamiliar with weighing the operational and security risks associated with important questions like, ‘Should my deployment be sovereign?’ or ‘How will we handle continuity of operations if there is a model provider outage?’
Forget the vendor’s feature list. Enterprises need to think about what they want first, then figure out if the vendor list matches. Otherwise, it’ll be back to the same dissatisfaction with SIEMs, UBA, and SOAR platforms. Some vendors recognize this, and smart enterprises can work with these committed vendors to achieve great success.
DJ: Will AI-powered security capabilities be enough to keep pace with increasingly sophisticated cyber threats?
Merza: No, and it’s not because the capabilities themselves are fundamentally weak. Security operations have always been a people, product, and process capability. Product features alone are insufficient. The biggest risk is with security teams’ learning curve. We’ve all seen this movie before. It happened with high-speed internet connectivity and then it happened with cloud computing. Security teams need time to catch up with the tech and its ramifications.
At the same time, this doesn’t mean security teams have to understand every detail. In modern deployments, most security analysts can’t explain how SSO works or how to reverse engineer malware. Those capabilities are out of reach for most organizations. But there is a base set of knowledge that high-performing teams have. Understanding AI-driven attacks and effectively using AI agents for security requires some base knowledge. Just like the cloud era, we can be sure that most in the security industry (buyers, users, and sellers) don’t know the kind of threats that are around the corner. Attackers have a learning curve too.
DJ: What are the biggest operational challenges enterprises face when implementing AI-powered SOC capabilities?
Merza: It’s in the name. AI systems are probabilistic. Applying CISSP 101 patterns is the single biggest problem organizations face. It’s better to start with outcomes, then go to processes, and then move on to the mix of people required and the tech those people want to use to address those outcomes. Too many security leaders are treating the AI moment as a tech problem. Yes, there is a tech challenge here, but operational resilience needs more than tech. One of the most important outcome-driven questions to ask is, ‘How much of what kind of work can be done by AI capabilities versus humans?’ Most people don’t have a pattern to answer that question. Surprisingly, most people still want humans in the loop for the most basic reasoning tasks.
Next comes hanging on to CISSP 101 best practices that are simply operational nonsense. For example, putting all your data in one place, normalizing your data across data lakes, or having an alert escalation process that is flawless. I can’t tell you how many people I’ve heard say, ‘Garbage in garbage out’. Yes, within bounds that’s true. But failed enterprise AI and big data projects are littered with the bodies of data-normalizing consultants.Next comes the operational part of it all. Most enterprises are struggling with gatekeeping AI access. The security teams have the opposite problem. They have to safeguard their own use of AI as they analyze questionable links, permissions, and files. They don’t have many tools to determine if their use of a model or AI harness will help them achieve the outcome, leak the data, or worse, execute the attackers intent.”
DJ: How will AI reshape the future of the enterprise SOC, particularly for smaller security teams and managed security service providers?
Merza: AI will democratize, but not in the way most believe. Overall, security products will improve. Therefore, AI will help accelerate teams. Most teams will be bound by the same things that hold them back today: their people, their processes, and a distant third, the products they use. The analogue is internet access and computers for every modern worker. Yet not every company is awesome, and not every security team is amazing.
What this means for security teams is that they’ll want to quickly determine who they are: builders, early adopters, or mainstream buyers. The builders and early adopters have a tremendous opportunity. They will shape the future by partnering with like-minded emerging tech companies who share design principles. For critical infrastructure, those are sovereignty, extensibility, and transparency. They will be open to challenging conventional processes and team tiering to reduce barriers around tool and code competency and security domain knowledge.
These teams will graduate from the dashboard-driven ops model to an integrated decision engine model. These teams won’t be driving dashboards. They will be operators of swarms of agents. Deep knowledge of security principles, coding principles, and IT infrastructure will be the number one resume qualifications for these operators.
The successful MSPs will also have to evolve their workforce. Customers will demand a lot from MSSPs as the commoditized tooling will handle much of the tech workload provided by MSSPs today. MSSP will become more valuable for their core value of offsetting risk.
DJ: How are enterprise customers evaluating AI-powered security solutions, and what matters most to them?
Merza: One Fortune 50 leader told me, ‘I don’t care if it has AI, ZI, or mice under the hood. I want outcomes for my spending.’
Critical infrastructure organizations operate in highly demanding environments with very savvy teams. Many of them are experimenting with building their own AI solutions. Their questions have gone beyond the 2023 hallucination controls to operational considerations on sovereignty, extensibility, API interfaces, model fail over, scalability and operational continuity.
The questions they’re asking are practical. Where does our data go? Can we choose which model to use? Can we inspect the reasoning behind an investigation? Will it work across the systems we already have? What happens to the cost as our investigation volume increases? They want evidence of operational improvement. They don’t want sales pitches.
Click Here For The Original Source.
——————————————————–
