# Security Overview

Treasure AI Studio is a multi-tenant AI agent execution platform designed with security at every layer. This page covers the architecture, data protection measures, execution environment, and monitoring capabilities that protect your data.

Note
The AI executes all commands and queries inside a sandboxed execution environment with restricted network access and filesystem scope. Your Treasure AI credentials are managed server-side and are not stored in the browser.

## Objective

Understand the security boundaries that protect your data when using Treasure AI Studio — the architecture, data protection, server-side sandbox, external service connections, network audit logging, and your responsibilities when working with AI-generated output.

## Prerequisites

- Signed in to Treasure AI Studio ([Getting Started](/products/ai-studio/getting-started))
- **Admin role** required for Network Audit Log and Network Policy access


## Architecture and Data Protection

### Platform Architecture

Treasure AI Studio operates as an AI orchestration layer hosted on AWS. Key architectural points:

- **Multi-tenant platform.** Each organization's data is logically isolated from other tenants.
- **CDP integration.** Studio accesses customer data through CDP APIs — it does not connect directly to your underlying data stores.
- **Server-side execution.** All agent operations run in isolated server environments, not on your local machine.


### Encryption

All data is encrypted both in transit and at rest:

| Protection  | Details |
|  --- | --- |
| **In transit** | TLS 1.2 or higher for all connections |
| **At rest** | AWS-managed encryption keys |


### LLM Inference and Data Privacy

Studio uses Amazon Bedrock for LLM inference with the following data protections:

| Protection  | Details |
|  --- | --- |
| **No model training** | Amazon Bedrock does not use customer data for model training. Your prompts, queries, and data are not used to train or improve foundation models. |
| **Zero data retention** | Amazon Bedrock does not retain prompts or responses after inference completes. Your data is not stored by the LLM provider. Note: AI Studio retains [chat history](/products/ai-studio/security/data-retention) based on your organization's retention policy. |


### LLM Data Residency

LLM inference requests are processed within a geographic boundary based on your Treasure AI account region:

| Account Region  | Primary AWS Region | Geographic Boundary |
|  --- | --- | --- |
| **US** | us-east-1 (N. Virginia) | US regions only |
| **EU01** | eu-central-1 (Frankfurt) | EU/UK regions only |
| **Tokyo** | ap-northeast-1 (Tokyo) | Japan regions only |
| **AP02** | ap-northeast-2 (Seoul) | Korea regions only |


Amazon Bedrock Cross-Region Inference is used for throughput optimization. Geographic inference profiles ensure that customer prompts and model responses **never leave the geographic boundary**. All inter-region data transmission remains on the AWS private network, encrypted in transit.

### User Data Isolation

All user data — including chats and files — is isolated per user. Administrators cannot access other users' chats or files.

## Server-Side Execution Environment

Unlike Treasure Studio Labs — which ran commands directly on your local machine — Treasure AI Studio executes all CLI commands and queries in an isolated server-side sandbox.

This means:

- **No local filesystem access.** The AI cannot read or write files on your computer. File operations happen in a managed server environment.
- **Sandboxed execution.** Each session runs in its own isolated worker with restricted network and filesystem access. If a sandbox crashes unexpectedly, you'll see the message: "The sandbox environment crashed unexpectedly and is restarting. Please wait a moment and try again."
- **Credential isolation.** Your Treasure AI credentials are managed server-side and are not stored in the browser. The AI accesses your Treasure AI account through a server-side access token — you do not need to provide credentials directly.


## External Service Connections

Studio agents can connect to external services (such as Google Drive, Slack, or other third-party tools) to retrieve data or perform actions.

### OAuth-Based Authorization

- Users authorize each external service connection explicitly through OAuth.
- Credentials are encrypted at rest and scoped to the individual user.
- Users can revoke connections at any time through Settings.


### Network Policy (Admin Only)

External connectivity is controlled by a **deny-by-default** network policy. Administrators can:

- **Allow specific domains.** Configure an allowlist of permitted external domains.
- **Block all external access.** Disable external connectivity entirely for the organization.


Note
Only Administrators can configure the Network Policy. Regular users cannot modify network access rules.

## Network Audit Log (Admin Only)

Account administrators can monitor all network activity through the **Network Audit Log** in Settings.

### How to Access

1. Open **Settings** (click your user avatar in the sidebar footer)
2. In the left navigation, under the **Organization** section, click **Network Audit Log**


### Audit Log Features

| Feature | Description |
|  --- | --- |
| **Time range** | Filter by duration: 1h, 6h, 24h, 3d, 7d, 14d, 30d |
| **Sort order** | Newest first or Oldest first |
| **Filters** | Domain, IP, Port, Chat ID, Action type ("Any" for all) |
| **Columns** | Timestamp, Chat, Action, Domain, IP, Port, Protocol |
| **Stats** | Shows matched count, scanned count, and bytes scanned |


Network Audit Log showing time range selector, filter controls, and activity table
Note
The Network Audit Log tab is only visible to account administrators. Regular users will not see the "Organization" section in Settings.

What Is Logged
The audit log records the **domain** of each external connection, not full URLs. Query parameters, paths, and request bodies are not captured.

## AI-Generated Output Disclaimer

Studio uses AI to generate outputs including queries, configurations, segments, and recommendations. These outputs:

- **May be incorrect or incomplete.** Always review generated SQL, segment rules, and configurations before deploying to production.
- **Should be validated.** Use the `sql-skills:trino-optimizer` skill to check query performance, ask the AI to run `tdx sg validate` to validate segment definitions, and `tdx journey validate` for journeys.
- **Are your responsibility.** Review and approve all AI-generated changes before pushing them to your Treasure AI account.


## Best Practices

### For Administrators Rolling Out to Teams

1. **Use work folders to enforce guardrails.** Create work folders with instructions like "Always validate before pushing" and "Never modify production segments directly."
2. **Monitor via Network Audit Log.** Review network activity regularly, filtering by time range and domain to identify unexpected API calls.
3. **Configure Network Policy.** Restrict external connectivity to only the domains your organization needs.


### For Individual Users

1. **Review tool calls.** Every action the AI takes is visible in the chat stream. Watch for queries and API calls as they execute.
2. **Leverage skills for validation.** Skills like `validate-segment` and `validate-journey` catch configuration errors before they reach your account.
3. **Manage external connections.** Only authorize external services you actively need, and revoke unused connections.


## Security Responsibility

You are responsible for:

- Reviewing and validating all AI-generated outputs before deployment
- Not embedding secrets, passwords, or access tokens directly into chat messages
- Reporting any security concerns to [security@treasure-data.com](mailto:security@treasure-data.com)


Note
Penetration testing, vulnerability scanning, or other security assessments of Treasure AI Studio require Treasure AI's prior written consent.

## Verification

- [ ] (Admins) Access the **Network Audit Log** under Settings > Organization
- [ ] (Admins) Review **Network Policy** settings for external connectivity


## Troubleshooting

| Issue | Solution |
|  --- | --- |
| "The sandbox environment crashed unexpectedly and is restarting" | Wait a moment and retry your message. If persistent, start a new chat — the operation may exceed sandbox resource limits |
| Audit log shows no entries | Audit log access requires admin permissions. Confirm your account role with your organization administrator |


## Next Steps

- [Data Retention](/products/ai-studio/security/data-retention) — Chat history retention and data lifecycle
- [Access Control](/products/ai-studio/security/access-control) — PBP, IP allowlists, and AI feature access
- [Getting Started](/products/ai-studio/getting-started) — Set up Studio on your platform
- [SSO Login](/products/ai-studio/security/sso-login) — Authentication details
- [Core Concepts](/products/ai-studio/concepts) — Work Folders, models, and credits