Securing Autonomous AI Agents With Google Cloud

Welcome to our eight-part defence-in-depth series for securing autonomous AI agents.

This series is designed to help you understand the real risk exposure and how to implement practical controls across data, identity, runtime and security operations. Each of these short sessions provides the technical depth and architectural context needed to get started with confidence.

How to use this series

New to agentic AI security? Start at episode one, and follow the lifecycle end-to-end
Solving a specific problem? Jump to data, identity, runtime, detection or SecOps
Sharing internally? Use episode summaries below to route the right session to the right team

Episode 1: Introduction

Discover the opportunity behind agentic AI, the security risks of autonomous digital workforces and why layered defence-in-depth is essential. Get started with a preview of what to expect throughout this series.

Episode 2: Risk profile

We’ll examine how agent autonomy expands the attack surface through access to your ‘crown jewels’. You’ll also explore some practical failure scenarios and see why these issues become board-level incidents.

Episode 3: Data hygiene and sanitisation

Learn why secure agents depend on secure data. We’ll look at how to use Wiz Cloud to gain visibility across the AI estate and Google Cloud Sensitive Data Protection to redact or tokenise sensitive information before agent access.

Episode 4: Data governance

Unstructured data becomes a ‘data swamp’ that drives unauthorised access. See how the Dataplex Universal Catalogsupports a governed data mesh, and how virtual private cloud (VPC) service controls can reduce exfiltration risk.

Episode 5: Identity and access

In this episode, we’ll look at how to move away from permanent credentials like hard-coded API keys. Learn how to useworkload identity federation for short-lived access, secret managers for legacy systems and least privilege practices through identity and access management (IAM) to limit blast radius.

Episode 6: Runtime protection

Securing the environments where agents execute, process data and interact with users is critical. We’ll discuss using confidential computing for memory encryption and Vertex AI guardrails to reduce prompt injection, toxic language and real-time data leakage.

Episode 7: Threat detection and response

No prevention layer is perfect, but active surveillance helps to close the gap. See how Google Security Operations (SecOps) correlates signals into high-fidelity threat stories and uses SOAR playbooks to accelerate response.

Episode 8: Efficient SecOps teams

We'll end by learning how to reduce operator burnout and make security expertise more accessible. With Wiz Cloud, wecan reduce alert noise, while Google SecOps supports rapid investigations, and Gemini can help analysts use natural language for threat hunting.

Securing AI Agents: A Defence-in-Depth Video Series