How Can We Prevent AI Data Leakage?

The CData AI Gateway provides a central control layer for an organisation’s AI usage, bringing AI access, data security, governance, cost control, and monitoring into one place.

Connect AI already provides controls such as roles and permissions, Toolkits, MCP restrictions, Workspaces, Data Security rules, and audit/query logs to control what AI agents can access and how sensitive data is handled.

Documentation: Getting Started - CData

The AI Gateway builds on these capabilities by providing a central control layer for how AI applications and agents interact with LLM providers and data. Instead of allowing applications to connect directly to different providers and data sources, the Gateway brings access, security, cost management, and monitoring into one place. This helps keep access controlled, credentials protected, spending monitored, and AI activity auditable.

The main components are:

1. LLM Providers (LLM Providers - CData)

The LLM Providers documentation describes how organisations can centrally register providers such as Anthropic, OpenAI, and Azure OpenAI. Provider API keys are stored securely and are not exposed directly to applications or users. Administrators can also control which models are available and define their associated costs, making it possible to track and manage AI spending.
 

2. LLM Gateway (LLM Gateway - CData)

The LLM Gateway documentation describes a single OpenAI- and Anthropic-compatible endpoint that applications can use instead of connecting directly to individual AI providers.

It provides several controls:

• Projects: Set monthly spending budgets and request rate limits.
• Routing Rules: Decide which provider and model should process a request. Routing can also support automatic failover and optional prompt enrichment.
• Virtual Keys: Issue credentials to applications that can be revoked when required.
• Guardrails: Inspect prompts for potential issues, such as exposed credentials or prompt injection attempts, before requests are sent to the LLM provider.

3. MCP Gateway (MCP Gateway - CData)

The MCP Gateway covers the central endpoint that allows AI agents to access tools and data.
Through Tool Servers, organisations can bring together CData connections and third-party MCP servers behind a single endpoint. Agents can authenticate using the organisation's SSO, while access to the underlying connections is handled securely for each request.

This covers the primary functionality offered by the AI Gateway. However, there is another important and closely related component that I would like to highlight: the Context Editor.
 

Context Editor (Context Editor - CData)

The Context Editor acts as a central repository where administrators and data owners can define datasets, metrics, and business rules using plain language. This context can then be provided to AI agents through a search tool and added to LLM prompts by the gateway.

The key advantage here is that AI interactions can consistently use the organisation’s own definitions, terminology, and business context. This context is maintained independently of the LLM provider, so it does not get lost if an organisation or user switches from one AI provider to another.

Overall
Taken together, these capabilities provide a central governance layer for AI. Connect AI focuses on controlling how AI clients access organisational data, including permissions, MCP tools, sensitive-data handling, and activity monitoring. The AI Gateway extends that control to the broader AI environment by managing LLM providers, model access, spending, routing, application credentials, MCP access, and security guardrails.