Two views, one SOC. Real world insights from both a CISO & an engineer on building trust with agentic AI

Agentic AI is moving rapidly from experimentation into active use within security operations. Across modern SOCs, AI-driven agents are increasingly being used to triage alerts, enrich investigations, recommend response actions, and in some cases autonomously execute elements of incident response. From a CISO perspective, this evolution presents a rare opportunity to materially improve cyber defence outcomes and increase efficiencies but also introduces a new category of operational, governance, and accountability risk.

Drawing on her experience as a CISO responsible for five operational Security Operations Centres across the APJ and MEA region, Kylie Watson will open the session with a candid, operations-focused perspective on what it actually takes to run secure operations as agentic AI becomes embedded in day-to-day SOC workflows. Rather than focusing on AI capabilities in isolation, her insights focus on the realities CISOs and SOC leaders are grappling with today: accountability for automated decisions, audit and regulatory scrutiny, resilience under failure conditions, and maintaining analyst trust as systems become increasingly autonomous.

She will outline how CISOs should view agentic AI systems not as tools, but as operational actors that require explicit trust boundaries, enforceable guardrails, and continuous oversight. Drawing on real-world examples from managed detection and response, incident response, and large-scale SOC operations, she will discuss where organisations are seeing genuine operational benefit and where poorly designed agents have introduced unacceptable risk and governance gaps.

A central theme of the opening section is accountability. From a CISO standpoint, "the AI did it" is never an acceptable answer. Kylie will examine the non-negotiable requirements SOC leaders should place on agentic systems to support executive accountability, audit, and regulatory review, including transparency of actions, evidentiary traceability, and operational explainability.

The session then transitions to Malak, who will take these CISO level requirements and translate them into practical engineering and architectural patterns for designing and running secure agentic AI systems inside the SOC. Malak will go deep into how these principles are implemented in practice, covering topics such as deployment models and trust boundaries for agentic workloads, guardrail enforcement mechanisms, and techniques to constrain agent behaviour while still enabling meaningful automation.

Malak will explore concrete strategies for mitigating hallucinations through retrieval-augmented approaches, and for governing agent memory and context to ensure decisions remain defensible and repeatable in operational environments. A strong focus will be placed on observability by design, including how to instrument agentic systems for full execution logging, structured outputs, and deterministic workflows that enable replay, validation, and post-incident analysis.

Resilience in AI-driven security operations will also be addressed at an engineering level. Malak will walk through failure-aware design patterns, safe degradation strategies, and human-in-the-loop controls that ensure agentic systems fail transparently rather than silently, preserving operator trust and operational safety during degraded conditions or unexpected behaviour.

The session provides both the executive lens and the engineering depth required to run secure, accountable, and resilient AI powered SOCs. Attendees will leave with a clear understanding of how to adopt agentic AI in security operations without surrendering control, accountability, or trust.