Brinqa AI Feature — Security & Data Privacy Documentation
Version: 1.0 Last Updated: April 8, 2026 Classification: Customer-Facing
1. Overview
Brinqa offers an AI-powered feature that allows customers to train models on their own data to automatically predict and fill in missing attribute values. This document describes the security architecture, data handling practices, and privacy controls that protect your data throughout the AI training, inference, and integration lifecycle.
The system has been designed with a foundational principle: your data never leaves your isolated environment, and the AI can only access what you already have access to.
2. Architecture Overview
2.1 High-Level Data Flow
2.2 Environment Isolation
3. Security Architecture
3.1 Tenant Isolation
Every customer operates within a fully segregated environment. This isolation is the foundation of Brinqa's security model and ensures the following guarantees:
- No cross-tenant data access. AI workloads for your organization run exclusively within your environment. There is no shared infrastructure, storage, or network path that would allow data to flow between customer environments.
- Credential-scoped access. The AI uses your user credentials to access data, meaning it can only reach data you are already authorized to view. No elevated or service-level privileges are granted to the AI beyond your existing access scope.
- No outbound data transfer. The AI cannot transmit data outside of your environment. There is no mechanism for sending your data to another customer's environment or to any external endpoint.
3.2 Ephemeral Compute (Kubernetes Pods)
AI training and inference workloads run in ephemeral Kubernetes pods that are purpose-built for a single operation and torn down afterward.
- Training pods pull data using your credentials, train the model, upload the trained model to your cloud store, and delete intermediary data.
- Inference pods pull data using your credentials, run predictions using your trained model, upload results via the Brinqa connector to your database, and delete intermediary data.
- No persistent state. Pods do not retain any data between triggered runs. Data is only retained between attempted retries on failed runs. Each invocation is a clean, isolated execution.
3.3 Credential Handling
The AI feature operates under your existing authentication and authorization framework. The system uses your user credentials to access data, which means:
- Access is limited to data the authenticated user is already permitted to see.
- No additional service accounts or elevated privileges are introduced by the AI feature.
- Credential management follows existing policies and controls.
4. Data Handling & Lifecycle
4.1 Data in Transit
During both training and inference, data moves between your cloud data store and the ephemeral Kubernetes pod. All data in transit is encrypted and travels within your isolated environment boundary.
4.2 Intermediary Data Files
The AI process generates intermediary data files during training and inference (e.g., preprocessed feature sets, temporary working files). These files are automatically and permanently deleted every time the pod finishes its operation. No intermediary data persists beyond the lifecycle of a single pod execution.
4.3 Model Artifacts
The only persistent artifact produced by the AI feature is the trained model itself. This model is stored in your cloud data store within your isolated environment. The model does not contain raw customer data — it contains learned statistical patterns derived from your data.
4.4 AI-Predicted Data & Human Approval
When the trained model generates predictions (filling in missing attribute values), the predicted data is uploaded to your database through the Brinqa connector. Critically:
- All AI-predicted attributes are flagged as AI-generated. This provides full transparency and traceability, so you always know which values were produced by the AI versus entered by a human or ingested from a source system.
- AI-predicted values are excluded from future training data until explicitly approved by a user. This prevents unapproved or potentially inaccurate predictions from compounding over successive training cycles.
- User approval is required before any AI-generated data is incorporated into the dataset used to train future models. This keeps you in full control of your data quality and the AI's learning foundation.
5. Security Controls Summary
| Control | Description |
|---|---|
| Tenant Isolation | Customer environments are fully segregated with no cross-tenant access |
| Credential-Scoped Access | AI uses user credentials; cannot access data beyond user's existing permissions |
| Ephemeral Compute | Kubernetes pods are created per-operation and destroyed after completion |
| Intermediary Data Deletion | All temporary data files are deleted after every model operation |
| No Outbound Data Transfer | Data cannot leave the customer's isolated environment |
| AI Attribution Flagging | All AI-predicted values are clearly marked as AI-generated |
| Human-in-the-Loop Approval | AI predictions require user approval before entering future training sets |
| Model-Only Persistence | Only trained models are stored; no raw data is retained |
6. Frequently Asked Questions
Can Brinqa employees see my data during AI processing? The AI operates within your isolated environment using your credentials. The processing is automated and does not involve Brinqa personnel accessing your data.
Could AI training on my data affect another customer's results? No. Environments are fully segregated. Your model is trained exclusively on your data and stored in your environment. No data or model artifacts are shared across customers.
What happens if I don't approve an AI-predicted value? The value remains flagged as AI-generated and is excluded from future training data. You can reject it, correct it manually, or leave it as-is. Unapproved predictions will not influence future model training.
Does the AI create copies of my data? During processing, temporary working files are created within the ephemeral pod. These are automatically deleted when the operation finishes. The only lasting artifact is the trained model, which contains statistical patterns rather than raw data.
Where is my trained model stored? Your model is uploaded to your cloud data store within your isolated environment, alongside your existing data. It does not leave your environment.
For questions about this documentation or the Brinqa AI feature's security architecture, please contact your Brinqa account representative or reach out to Brinqa's security team.