Prepare and test your skills
Prepare and test your skills
Security and governance of AI workflows in Gemini Enterprise (opens in a new tab) is the framework that controls who can access enterprise data, what agents can do with it, and how activity is monitored. Gemini Enterprise is an intranet search, AI assistant, and agentic platform that connects Google and third-party data sources to provide permissions-aware search, grounded answers, and autonomous agents. Governance combines built-in trust controls with administrator configuration so sensitive assets stay protected while workflows automate multi-step tasks.
Permissions-aware access in Gemini Enterprise means search results and generative answers respect each user's existing permissions at scale. Data flows from connected data sources into dedicated data stores, then through a Gemini Enterprise app (opens in a new tab) to the end user, and the app only serves content the authenticated user is allowed to see. Access is established through single sign-on (SSO) and user-level access, which provides built-in trust across cloud platforms, CRMs, productivity tools, and legacy systems.
Gemini Notebook Enterprise (opens in a new tab) applies a narrower trust boundary for curated research. A notebook is an organized collection of hand-selected documents, websites, and sources, and chat-based Q&A responses are strictly confined to that notebook's information. Data in notebooks remains under access control and is not used for external model training, so teams can synthesize insights and generate FAQs, timelines, and audio summaries without exposing broader enterprise data.
Identity and Access Management (IAM) in Gemini Enterprise controls which users, agents, and service accounts can view data, invoke agents, and change configuration. Administrators deploy and manage the platform by configuring security policies, assigning user licenses, connecting enterprise data sources, and managing agent service accounts and permissions. Policy enforcement in Agent Platform applies IAM policies alongside Model Armor policies to govern agent behavior from deployment onward.
Authentication to Gemini Enterprise supports three peer frameworks:
Workforce pool administration carries elevated risk because the IAM Workforce Identity Pool Admin (roles/iam.workforcePoolAdmin) and IAM Workforce Pool Editor (roles/iam.workforcePoolEditor) roles can modify workforce pool attribute mappings. Incorrect changes can enable user impersonation, which can lead to unauthorized document access and agent actions. To reduce this risk, grant these roles only to trusted administrators who absolutely require them, use Privileged Access Manager to grant time-bound entitlements and audit their use, and integrate Workforce Identity Federation audit logs into a security monitoring solution such as a SIEM.
Exam tip: Use Workforce Identity Federation when sending data with a third-party connector, and restrict roles/iam.workforcePoolAdmin and roles/iam.workforcePoolEditor to trusted admins with Privileged Access Manager entitlements.
Data security in Gemini Enterprise protects enterprise content from threats, breaches, and identity theft across ingestion, storage, and generation. Gemini Enterprise is integrated with VPC Service Controls (VPC-SC) (opens in a new tab), which places a service perimeter around supported Google Cloud resources to control data movement. Data is encrypted by default with Customer Managed Encryption Keys (CMEK), with support for External Key Manager (EKM) or hardware security module (HSM) keys subject to Cloud Key Management Service limitations.
Privacy protection extends to content filtering and origin transparency for agent outputs. A multi-layered approach combines configurable filters, system instructions, data loss prevention (DLP), and Gemini as a filter to protect against content safety violations, brand safety issues, model drift, and hallucinations across text, images, video, and audio. For supported models, Agent Platform automatically adds cryptographically signed C2PA Content Credentials to generated images to indicate AI origin and verifiable history, and user-requested data is deleted within 60 days under Google Cloud deletion controls.
Compliance and observability in Gemini Enterprise provide transparency into where data resides, who accessed it, and how agents behaved. Compliance resources include access transparency, audit logging, supported locations, and compliance certifications and security controls. Audit logs and usage audit logs track agent activities and system activity, while end-to-end request traces and trace spans show the execution path of search and agent workflows.
The Cloud Observability suite collects telemetry by default when agents are deployed with the Agent Development Kit (ADK). Telemetry flows from Agent Platform, Agent Gateway, and Model Armor to Cloud Trace, Cloud Logging, Cloud Monitoring, and Topology using OpenTelemetry protocols. Traces capture execution paths, logs capture events and errors, and metrics capture latency and token usage, so administrators can debug failures, monitor costs, trace agent reasoning, and view application topology. The GenAI Evaluation Service supports ongoing online evaluation of agent quality, and continuous safety evaluation covers content safety, brand safety, relevance, bias and fairness, truthfulness, and robustness to adversarial attacks.
Agent governance is divided across three admin surfaces that together control the full agent lifecycle from creation to production management. The following table compares what each surface governs and the key controls it provides:
| Admin surface | What it governs | Key controls |
|---|---|---|
| Agent Platform (opens in a new tab) | Central lifecycle governance for platform and security administrators | Agent Registry versioning and monitoring, Agent Gateway configuration, IAM and Model Armor policy enforcement, identity for agent service accounts, Security Command Center integration, and observability |
| Gemini Enterprise Admin | Agents and data within a Gemini Enterprise instance | Licenses and users, instances and data connectors, Agent Gateway routing, adding agents and tools from registries, user permissions for agents, and Cloud logs, metrics, and traces |
| Google Workspace Admin Console | Interaction between agents and Google Workspace services and data | Enabling the service for users, user permissions for agents, domain policies on data access, and Workspace audit logging for agent interactions |
App Design Center templates help enforce governance from the moment an agent is deployed, including a Simple Agent Platform template and an Agent Platform with governance template. Gemini Enterprise Admin directs administrators to Agent Platform for policy enforcement and additional observability, and the Workspace Admin Console directs administrators to Agent Platform for gateway configuration and comprehensive governance. This separation keeps broad policy and threat detection in Agent Platform while instance, license, and domain controls stay with their respective admins.
Knowledge synthesis with NotebookLM is the practice of turning scattered technical documents into a single trusted reference that can answer questions and create new materials. Gemini Notebook Enterprise (opens in a new tab) provides notebooks that hold curated, hand-selected documents, websites, and other sources on one topic, project, or department. Each notebook becomes the single reference point for deep understanding, content creation, and knowledge sharing from those trusted sources. Subscriptions for Gemini Enterprise (opens in a new tab) (Standard and Plus) include Gemini Notebook Enterprise, and Gemini Notebook Enterprise is also available as a standalone product.
Centralized notebooks in Gemini Notebook Enterprise are organized collections built only from sources you intentionally select and provide. Data flows from your uploaded documents, websites, and specific sources into the notebook, and all later answers and drafts flow back out from that same bounded set. The core function is understanding and extracting from unstructured text and generating content based on your sources, while your data remains under access control and is not used for external model training.
Targeted chat and synthesis in Gemini Notebook Enterprise let teams ask questions and generate usable artifacts without leaving their trusted sources. Chat provides question-and-answer responses strictly confined to the notebook information, so answers stay grounded in the selected documents. One-click generation transforms the same notebook content into new forms for sharing, review, and on-the-go consumption:
Gemini Notebook Enterprise fits cloud infrastructure work when the authoritative sources are already known and the goal is rapid, evidence-based synthesis. The notebook acts as the AI-powered reference point for a system, migration, project, or department, and every output can be traced to those curated inputs. Choose Gemini Notebook Enterprise when you want to:
Gemini Notebook Enterprise and Gemini Enterprise divide work by breadth of input and type of output. Gemini Notebook Enterprise focuses on focused knowledge management from your own documents, while Gemini Enterprise handles enterprise-wide search, multi-modal content generation, and agent orchestration.
| Product | Primary goal | Input | Output |
|---|---|---|---|
| Gemini Notebook Enterprise | Centralized knowledge and content generation from your own documents | Uploaded documents, websites, and specific sources | Synthesized insights, FAQs, timelines, interactive audio podcasts, and chat answers confined to sources |
| Gemini Enterprise | Enterprise-wide data search, multi-modal content generation, and agent orchestration | Broad queries and data from across enterprise systems, including Google and third-party SaaS | Answers from diverse data, text and multi-modal content, and automated actions by agents |
Exam tip: Choose Gemini Notebook Enterprise when answers must stay confined to a known set of sources; choose Gemini Enterprise when the starting point is unknown across enterprise systems or the task needs Excel, CSV, and images embedded within PDFs.
The joint workflow between Gemini Enterprise and Gemini Notebook Enterprise connects broad discovery to curated synthesis. Gemini Enterprise first searches across Google systems and third-party SaaS applications, which often leads to discovery of new, valuable sources. Users then add those newly discovered sources directly into a Gemini Notebook Enterprise notebook, which enriches the curated knowledge base. This sequence makes Gemini Enterprise the engine for discovery and Gemini Notebook Enterprise the system for trusted, source-confined understanding and content creation.
A four-step process flow showing Gemini Enterprise searching across Google and third-party SaaS systems, discovering new sources, adding them to a Gemini Notebook Enterprise notebook, and producing trusted, source-confined outputs.
Architecting and grounding AI agents in Gemini Enterprise is the process of building conversational agents that answer natural-language questions with accurate, enterprise-grounded results. A conversational data agent (opens in a new tab) combines a persona, a set of data sources, and business knowledge critical for its purpose. The agent translates a question into SQL, runs the query against the connected database, and then summarizes or visualizes the results for the user.
A conversational data agent is built from three connected parts where persona defines how the agent behaves, data sources define what data it can read, and business knowledge defines how it should interpret that data. Requests flow from the user in natural language to the agent, then from the agent to the database as generated SQL, and finally back to the user as an answer with summary or visualization. This design reduces complexity for application developers because users get insights from operational databases such as Spanner and AlloyDB for PostgreSQL (opens in a new tab) without learning query languages.
Agents support two lifecycle states that control who can use them. An agent starts as a draft for personal testing by its creator, and it moves to published when it is made available to other users in the project or through an API. Gemini Enterprise serves as the central hub where published agents created across Google Cloud applications can be found and managed. Choose draft state when tuning behavior and accuracy, and choose published state when the predefined context and analytics are ready for a broader audience.
The platform provides three core benefits for developers building these agents:
Grounding with authored context is the primary way an agent creator improves accuracy for business questions. Authored context includes schema descriptions, system instructions, and structured context that tells the model what tables and fields mean and what kinds of queries are expected. The agent depends on this context to choose the correct data source, generate valid SQL, and avoid guessing when names or business terms are ambiguous.
Before creating a grounded agent, the Google Cloud project must meet two dependencies. First, billing must be enabled for the project. Second, the project must enable the Data Analytics API with Gemini and the Gemini for Google Cloud APIs, plus the database API for the connected source, such as Spanner or AlloyDB for PostgreSQL. Enabling APIs requires the serviceusage.services.enable permission, which is included in the Owner role or the Service Usage Admin role.
Grounding with Google Search (GWGS) is an AI feature on the Gemini Enterprise Agent Platform that grounds companion agent suggestions in real-time Google Search results. Grounded results are generated from three inputs together: the end-user prompt, customer-provided context, and Google Search engine data, alongside recommended search suggestions. By using live web information, the companion agent (opens in a new tab) can give accurate, data-driven responses for current news, global knowledge, or other dynamic subjects.
GWGS creates a trust boundary tradeoff that architects must plan for. When this tool is enabled, search requests flow from the agent platform to Google Search outside the customer controls that normally protect enterprise data. GWGS does not support Customer-Managed Encryption Keys (CMEK), Data Residency (DRZ), Virtual Private Cloud Service Controls (VPC-SC), or Access Transparency (AXT). Use authored context grounding when answers must stay inside enterprise data controls, and use GWGS only when the need for current web facts outweighs those control requirements.
Tools in Gemini Enterprise for Customer Experience let virtual agents interact with other systems to retrieve, update, format, or analyze information. The agent depends on the tool as a connector: the model decides what action is needed, calls the tool, and then uses the returned data in its response. AI coach (opens in a new tab) can reuse these same tools to improve its suggestions to human customer service agents.
A tool is created once in the same project as the agent and then selected during instruction design:
To add the tool to AI coach, sign in to the Agent Assist console, open AI coach, type $ for the tool, and select one of the available tools from the list. The console lists only tools created in the same project, so the tool and the coach share a project boundary. This project scoping ensures the coach can only call tools that the owning team has explicitly created and managed.
Publishing moves an agent from private testing to shared use through an ordered handoff between Looker, Google Cloud, and Gemini Enterprise. First, a data agent editor with the publish_agent_externally permission publishes the agent in Looker to the Gemini Enterprise instance shown in the Admin panel. Then, a user with the Discovery Engine Admin role grants access to the agent in the Google Cloud console. Only after those steps can a Gemini Enterprise user with data agent access find and chat with the agent in the Gemini Enterprise instance.
Access to agents is controlled by both project roles and content sharing. The project roles separate creation, management, and use:
| Use case | Required role |
|---|---|
| Create agents in a project | Gemini Data Analytics Data Agent Creator, which also grants Owner on agents you create |
| Edit, share, or delete a specific agent | Gemini Data Analytics Data Agent Owner or Editor |
| Discover and chat with published agents | Gemini Data Analytics Data Agent User |
Sharing grants content access so other users can chat with the agent and its Explores, and revoking removes that access within a minute or two while changes propagate. Granting content access to an agent does not also grant access to the underlying Explore or model (opens in a new tab), so a Looker admin must still grant Conversational Analytics and model permissions.
Exam tip: Sharing an agent lets users chat with it, but users still need separate permissions for the Explore and model behind it.
Gemini Notebook Enterprise focuses on centralized knowledge and content generation from hand-selected documents, websites and specific sources, with chat answers strictly confined to that notebook information. Gemini Enterprise handles enterprise-wide search, multi-modal content generation and agent orchestration using broad queries and data from across Google and third-party systems. Gemini Notebook Enterprise outputs synthesized insights, FAQs, timelines, interactive audio podcasts and confined chat answers, while Gemini Enterprise outputs answers from diverse data, text and multi-modal content and automated actions by agents.
Choose Gemini Notebook Enterprise when authoritative sources are already known and the goal is rapid, evidence-based synthesis within a specific, limited set of documents. It is the right choice for establishing a single AI-powered reference point for a topic or project, creating chat-based question-and-answer experiences from defined sources, and generating study guides, FAQs, timelines and interactive audio formats from that curated knowledge base. Choose Gemini Enterprise instead when the starting point is unknown across enterprise systems or the task needs Excel, CSV and images embedded within PDFs.
Permissions-aware access in Gemini Enterprise means search results and generative answers respect each user existing permissions at scale. Data flows from connected data sources into dedicated data stores, then through a Gemini Enterprise app to the end user, and the app only serves content the authenticated user is allowed to see, with access established through single sign-on and user-level access.
Grounding with Google Search (GWGS) grounds companion agent suggestions in real-time Google Search results using the end-user prompt, customer-provided context and Google Search engine data. When enabled, search requests flow from the agent platform to Google Search outside the customer controls that normally protect enterprise data, and it does not support Customer Managed Encryption Keys (CMEK), Data Residency (DRZ), Virtual Private Cloud Service Controls (VPC-SC) or Access Transparency (AXT).
Your enterprise organization is deploying Gemini Enterprise to provide employees with conversational AI assistants and custom AI agents connected to sensitive internal knowledge repositories. Your security team mandates strict security and governance controls with the following requirements:
Which architectural solution should you implement to satisfy these security and governance requirements?
Establish Private Service Connect endpoints for Vertex AI APIs, apply an Organization Policy constraint for Resource Location Restriction, and assign employees the basic Viewer (roles/viewer) IAM role at the project level.
Configure a VPC Service Controls perimeter enclosing the Gemini Enterprise project, add the Discovery Engine API (discoveryengine.googleapis.com) as a restricted service, apply an Access Context Manager basic access level matching corporate IP subnetworks, and grant users the Discovery Engine User (roles/discoveryengine.user) IAM role.
Configure VPC egress firewall rules to restrict outbound traffic to private Google APIs, define IAM Conditions on the Cloud Assist User (roles/geminicloudassist.user) role, and restrict access using a Workforce Identity Federation pool.
Deploy Google Cloud Armor security policies with WAF rules on an External Application Load Balancer, add cloudaicompanion.googleapis.com to a VPC Service Controls perimeter, and grant users the Discovery Engine Admin (roles/discoveryengine.admin) IAM role.
Last updated