TLDR: Google unveiled a suite of new AI and cloud security capabilities at its Security Summit 2025, focusing on safeguarding AI agents, enhancing security operations, and strengthening overall cloud defenses. Key announcements include advanced protections for Google Agentspace and Agent Builder, a new AI-powered alert investigation agent, and significant updates to Security Command Center and Identity and Access Management.
Google today showcased a comprehensive array of new artificial intelligence and cloud security features at its annual Security Summit 2025, held on August 19. The announcements underscore Google’s commitment to securing the rapidly evolving landscape of AI innovation and bolstering enterprise defenses against sophisticated cyber threats. The updates span critical areas including robust protections for AI agents, advanced tools for Security Operations Centers (SOCs), significant enhancements in core cloud security, and expanded safeguards for enterprise browsing.
AI Security Takes Center Stage
AI security was a prominent theme, with Google introducing new capabilities specifically designed for Google Agentspace and Agent Builder. These include expanded AI agent inventory and risk identification, advanced in-line protection and posture controls, and proactive threat detection and management. Naveed Makhani, Product Lead for Security AI at Google Cloud, articulated the company’s strategic approach: “Our philosophy is to make the secure path the easy path, without sacrificing the flexibility developers need. We provide a layered set of built-in protections: detective policies to give security teams visibility without restricting developers and preventative controls to block high-risk actions by default. For developers who need maximum control, our core security capabilities like Model Armor are available via a REST API, giving developers the flexibility to define tailored protections for their application’s specific use cases and associated risks.”
Google also elaborated on its vision for an ‘Agentic SOC,’ a system where AI agents collaborate to automate triage, investigation, and response workflows. A new alert investigation agent, currently in preview, is designed to autonomously enrich events, analyze command-line inputs, and construct process trees, leveraging methodologies derived from Mandiant’s frontline analysts. Payal Chakravarty, Director of Product Management for Google Security Operations, stated that this approach is intended to fundamentally alter how human security analysts operate, augmenting their capabilities.
Further enhancements to Model Armor, Google Cloud’s in-line protection system, now provide real-time safeguards for Google Agentspace prompts and responses, defending against threats such as prompt injection, jailbreaking, and sensitive data leakage. Google-owned Mandiant is also expanding its advisory role in AI security, offering new services for governance frameworks, AI environment hardening, and threat modeling. These services aim to assist organizations in assessing risks through adversarial testing and implementing robust defense mechanisms prior to deploying generative or agentic AI systems.
Expanding Cloud Security Tools
Also Read:
- Google Deepens AI Integration in Workspace, Bolstering Productivity for Businesses
- Google Champions Continuous Experimentation and Refinement in AI Agent Development
Beyond AI-specific advancements, Google introduced new features across its broader cloud security portfolio. Updates to the Security Command Center now include Compliance Manager, Data Security Posture Management, and Risk Reports, all aimed at streamlining compliance efforts and highlighting emerging threats. In the realm of identity and access management (IAM), Google previewed an IAM role picker powered by its Gemini AI, agentic IAM for automated provisioning of agent identities, and enhanced re-authentication mechanisms for sensitive actions.
