Organizations are deploying autonomous AI agents into production at scale: agents that authenticate, execute multi-step workflows, and make decisions across cloud infrastructure. These agents operate at machine speed, which means security operations must too. When agents behave unexpectedly, the pattern closely resembles an insider threat, but traditional detection and response cycles were designed for human-speed investigation. In this session, we'll share what we're seeing across enterprise environments as agentic workloads go into production: where the detection gaps are, why observability and security telemetry must converge at the application layer, and how organizations are building response capabilities that contain and remediate autonomously in seconds, not hours. We'll walk through practical patterns for calibrating automated response: when to contain, when to escalate, and when to let the system close the loop. The goal is to make security as fast and autonomous as the workloads it protects.
*Sponsored by AWS
The rapid shift to autonomous AI agents in the cloud is accelerating security concerns. Traditional, human-centric systems are overwhelmed by the millions of new digital identities agents create, exacerbating issues like compromised identities (the source of over 70% of cloud breaches) and sophisticated agent hijacking attacks. This talk argues for the urgent application of Zero Trust architecture to AI infrastructure. We present a hybrid framework that anchors on three core, non-negotiable principles for securing AI agents: Human Oversight for governance and critical actions, Innovation with Control through dynamically limited privileges, and Transparency via centralized logging and auditability.
*Sponsored by Google
Virtual
Agentic AI represents a structural shift in enterprise software, not an incremental improvement. For decades, organizations built security frameworks around a foundational assumption: humans make decisions, machines execute them. Agentic systems invert that model. They reason, plan, and act autonomously across identities, data stores, tools, and other agents, operating at machine speed in environments designed for human-paced decisions.
The security architecture most enterprises have in place was not built for this. Perimeter controls, identity governance frameworks, data classification policies, and audit mechanisms were designed for stateless, deterministic workloads. Agentic systems are stateful, adaptive, and capable of lateral tool invocation across an organization's most sensitive systems in a single workflow. The gap between existing controls and the actual risk surface of a production agentic deployment is not a configuration problem. It is an architectural one.
*Sponsored by Microsoft
Secure agentic deployment requires six foundational governance primitives, applicable regardless of platform or vendor: agent identification and inventory, data labeling and policy enforcement, agent security posture management, conditional access for agentic workflows, runtime threat detection and remediation, and compliance reporting. These primitives are not new products to be purchased. They are the deliberate extension of Zero Trust principles to a new class of non-human actor, and they must be architected into agentic systems from day one rather than retrofitted after the first incident. The organizations that adopt agentic AI safely will not be the ones with the most policies. They will be the ones that treat identity as the foundational control plane, govern the data agentic systems can reach before they govern what those systems generate, and build continuous adversarial testing into their security posture from the start. That window is open now, and it is the defining cloud security decision of this generation.
Virtual
As autonomous AI agents move into production across cloud environments, security teams face a fundamental mismatch: agents authenticate, execute, and make decisions at machine speed, while many security architectures were designed for human-paced, deterministic workloads.
Frank Kim leads a discussion with security leaders from AWS, Google Cloud, and Microsoft on the realities of securing AI-driven cloud environments. Marking the fifth year of Cloud Security Exchange, the panel will explore how cloud security has evolved over the past five years and what comes next as organizations embrace increasingly autonomous systems.
