How Actian AI Analyst protects your data¶
This page describes Actian SaaS, the service Actian hosts. In a self-managed deployment, your organization operates the infrastructure (database, storage, network, TLS certificates, backups and encryption keys) and the operational controls below apply to Actian's development and release process rather than to your runtime. See Self-managed deployments.
For a detailed, step-by-step account of what AI Analyst stores, sends to AI providers and deletes, see How AI Analyst handles your data.
Key security considerations¶
Security is covered at three levels below: design, technical measures and operations. For Actian SaaS, these commitments are set out in your agreement and Data Processing Agreement (DPA). Services you choose to connect (such as your own LLM endpoint) are governed by your agreement with that provider. For a detailed overview of our security mechanisms, see the Actian AI Analyst Security Solution Brief.
Security by design¶
AI Analyst is built to process only the data it needs. For example, data source credentials are encrypted before they are stored and never shown again after saving. Our engineering team is trained in current security practices. Actian SaaS follows practices such as least-privilege access, logical separation and infrastructure as code.
AI Analyst works from your semantic layer's definitions and retrieves only the rows a question needs; it does not replicate your databases. Query results are kept with the conversation and its artifacts, and CSV files you upload are stored so they can be queried.
Technical security¶
Technical measures include:
- Encryption in transit and at rest: TLS for connections to your data sources (you choose the SSL mode on each connection), AES-256-GCM encryption for data source credentials and AI endpoint API keys, and at-rest encryption of stored data by the hosting platform.
- Data integrity checks using signatures and hashes.
- Signed container images and software bills of materials (SBOMs).
- Regular system and software updates.
- Firewalls.
- Internal virtual networks.
- Pattern-based masking of common personal data (email addresses, phone numbers, US Social Security numbers, credit card numbers and IBANs) and replacement of result-row fields in diagnostic traces, where tracing is enabled.
- Separation between production and development environments.
- Separation of storage and processing systems.
- Tenant data separation: each organization's data is kept apart, and Workspaces inside an organization are a hard boundary.
- Logging and audit trails.
This is not an exhaustive list. For detailed information, refer to your Data Processing Agreement.
Access and identity controls¶
AI Analyst gives administrators these controls over who can reach your data:
| Control | What it does | Where |
|---|---|---|
| Capabilities | Each member's access is a set of capabilities: for the organization (for example Manage organization, Manage members and access, Manage inference) and per Workspace (Chat, Edit, Manage members). | Settings → Members |
| Workspaces | Nothing in one Workspace (semantic layer, AI Analysts, conversations, artifacts) is readable from another. A Workspace can own a private data connection. | Settings → Workspaces |
| Require MFA | Everyone who signs in with password, Google or Microsoft must enroll multi-factor authentication. | Settings → Administration → Security |
| Enterprise SSO and Require SSO | People sign in through your company's identity provider; Require SSO turns off other sign-in methods. | Settings → Administration → Security |
| SCIM provisioning | Your identity provider creates, blocks and deletes sign-in accounts. | Settings → Administration → Security |
| Restrict connection management | Only members with Manage organization can add, edit or delete data connections. Other Studio members keep read access. | Settings → Administration → Security |
| Audit | Records who did what, newest first, including actions Actian staff take on your organization. You can export it. | Settings → Environment → Audit |
Availability
- Who can do this: members with the Manage organization capability (security settings and audit) or Manage members and access (members and capabilities).
- Plan: Enterprise SSO is included only in plans that enable Enterprise SSO.
- Deployment: Actian SaaS and self-managed deployments.
See Security, Members, Workspaces and Audit Logs.
Data source protections¶
- Encrypted credentials. Credentials are encrypted at rest and never shown again after saving. They are decrypted only to connect to your data source and are never sent to the AI provider.
- Read-only access. AI Analyst only issues read queries. Some databases (PostgreSQL, MySQL, MariaDB, Amazon Redshift, MotherDuck and ClickHouse) also enforce read-only mode for the session. For the others, Studio shows "This Connector cannot enforce read-only access itself. Supply a warehouse user that is limited to read-only privileges." Every query is limited in rows and run time.
- Known source addresses. In Actian SaaS, the connection setup help lists the AI Analyst egress IP addresses, so you can allow only those through your database firewall.
- Restricted outbound calls. AI Analyst refuses to call addresses the deployment is not allowed to reach, for data source connections and for AI endpoints your organization configures.
AI controls¶
- Choice of AI provider. By default, Actian SaaS uses Actian Inference. If your plan includes customer-hosted inference, members with the Manage inference capability can point AI Analyst at your own OpenAI-compatible LLM and Embedding endpoints under Settings → Administration → Inference, so prompts and data go only to a provider you chose. Requests never fall back to Actian Inference.
- Steward approvals. By default, Steward asks for approval before it runs SQL against your data source or changes the semantic layer. Members with the Edit capability manage this under Steward's Tool permissions. This is a workflow safeguard, not a security boundary: members with Edit can change it to Always allow, and sessions can override it. To keep data out of reach, limit what the connection's warehouse user can read.
- Redacted diagnostic traces. Where diagnostic tracing is enabled, result-row fields are replaced by a placeholder and common personal-data patterns are masked before a trace leaves AI Analyst; values quoted in text are not removed. See Diagnostic traces.
Operational security¶
Actian also applies operational procedures to protect customer data, including:
- Physical and virtual access controls
- Strict access and account management protocols
- Device management
- Non-disclosure agreements with employees and subcontractors
- Regular security training for staff
- Logging and audit trails
When an organization is deleted, its data is permanently removed after a grace period. See When an organization is deleted.
For more details on our security posture, see the Actian AI Analyst Security Solution Brief.
Self-managed deployments¶
You can also run AI Analyst on your own infrastructure. In a self-managed deployment:
- You operate the runtime. Your organization provides and secures the PostgreSQL database, file storage, network, TLS certificates and backups.
- The same signed images. Actian ships the same container images, with signatures and SBOMs you can verify offline.
- Your encryption key. Data source credentials and AI endpoint API keys are encrypted with a key your operator generates and keeps in the installation's configuration.
- Your AI models. No AI model ships with the product; AI Analyst uses only the LLM and embedding endpoints your administrators configure.
- Air-gapped option. On an air-gapped deployment, the AI Analyst server makes no calls to Actian or to other vendors' services, and it refuses to start if a vendor-hosted service is configured. Bug reports stay on your installation until an administrator hands them to Actian; see What is shared with Actian.