# RudderAI Security and Compliance

This guide explains RudderAI’s security model, access controls, and data privacy policy.

Access control

Workspace isolation: RudderAI can only access your production workspace. It cannot access other workspaces.
Permissions: RudderAI has no write access to any workspace components. All operations are read-only.
Slack shared channel access: Anyone in the designated Slack channel can use RudderAI — this includes users who may not have direct access to the connected RudderStack workspace.
Dashboard access: Organization admins control dashboard access using the Enable RudderAI organization setting. On self-serve organizations that have access, this setting is on by default. See RudderAI in Dashboard for more information.

In Slack, all interactions are visible to everyone in the channel. RudderAI responds only in the designated shared channel, not in other channels or direct messages.

Revoke access

For dashboard access, organization admins can turn off the Enable RudderAI toggle in Settings > Organization. For Slack access, contact your RudderStack representative to revoke RudderAI access. Once revoked, RudderAI will be removed from your Slack channel and immediately lose access to your RudderStack workspace.

Data privacy

This section covers details on the model provider used for RudderAI and its data retention and PII handling policies.

AI model provider

RudderAI uses Amazon Bedrock for processing your queries and generating responses.

No data is used to train AI models, and conversation data is not stored by the AI provider.

PII handling

RudderAI provides comprehensive PII masking and protection. RudderAI has no access to unmasked customer PII.

Tools that access event payload data automatically mask sensitive fields before sending to the AI. Masking preserves data structure for debugging while protecting actual values.

Access control

Revoke access

Data privacy

AI model provider

PII handling

See more