Privacy is the whole point of the Gateway: it exists so you can use AI without handing sensitive data to a model you don't control — an AI gateway that redacts PII, PHI, and cardholder data before a prompt ever leaves your boundary. This page explains the guarantees and the controls behind them.
The core guarantees
- Sanitized before send — sensitive data is detected and redacted, masked, or blocked before any model call. Raw values never leave your boundary unless your policy explicitly allows it.
- No training on your data — prompts, documents, and responses processed by the Gateway are never used to train third-party models, and provider "no-training" terms are enforced at the connector.
- Response inspection — the model's reply is checked for leaked data on the way back, not just the prompt on the way out.
- Everything logged — a complete, exportable audit trail records what was sent, what was redacted or blocked, which policy version applied, and which model answered.
Data residency & deployment
| Option | Where data is processed |
|---|---|
| intSignal cloud | In the region you select. |
| Your VPC / on-prem | Inside your own network; the Gateway runs where your data lives. |
| Private model | With a self-hosted model, no third-party model is involved at all. |
Choose per project, so a regulated workload can run fully on-prem while a low-risk one uses the cloud.
Access & keys
- SSO via Microsoft Entra ID and role-based access control who can change policies, read logs, or manage keys.
- Secrets — model-provider credentials and Gateway keys are stored encrypted and never returned after creation.
- Least privilege — scope Gateway keys to a single project.
Audit & evidence
Every request produces an audit record you can search, retain to your policy, and export to a SIEM. When an auditor asks whether customer data reaches your AI provider, you can show exactly what was redacted on real traffic — evidence, not a questionnaire answer.
Compliance mappings
| Framework | How the Gateway helps |
|---|---|
| HIPAA | Redacts PHI before model calls; full access logging; on-prem option for ePHI. |
| PCI DSS | Masks or blocks PAN before any model sees it; no cardholder data in prompts. |
| GDPR | Data minimization by redaction; residency controls; no-training guarantee. |
| SOC 2 | Access control, audit trail, and change-managed, versioned policies. |
The Gateway is a control, not a substitute for policy
The Gateway enforces your rules extremely well, but you still own your data-handling policy, retention decisions, and provider agreements. Configure retention and access to match your obligations — and confirm your specific regulatory scope with your compliance team.
Related
- Define enforcement in Policies.
- Keep coverage current with Automatic policy setting.
- Pair with the Compliance Platform and Security Suite for a full posture.