Sensitive Data

Sensitive Data rules protect information at the employee prompt boundary. Before a supported agent submits a prompt, Gesta checks the content against active organization rules and applies the configured outcome.

Use Sensitive Data rules for:

• credentials and access tokens;
• private keys;
• customer identifiers;
• regulated or confidential data;
• internal data patterns that should not be sent to an AI service;
• organization-specific identifiers or secrets.

Rule outcomes

Choose the outcome according to the data category, destination, and organization policy. Use Block for content that must not cross the boundary.

Create a Sensitive Data rule

Open Security → Sensitive Data, then select New rule.

Configure:

• Name and Description;
• Status and responsible Owner;
• Source to inspect;
• Detector type and pattern;
• Category and Severity;
• Action: Record, Warn, or Block.

Design the detector

Start with a specific, testable pattern. A detector should identify the organization's sensitive value without matching common source code or ordinary conversation.

Recommended process:

1. define the data category and why it needs protection;
2. identify a stable format or marker;
3. create representative synthetic examples;
4. test both matching and non-matching samples;
5. begin with Record or Warn when false positives are uncertain;
6. review findings before moving to Block.

Do not test with real secrets or customer data.

Review findings

The Sensitive Data page shows rule status, findings, categories, severity, and outcomes. Open a finding to understand:

• which rule matched;
• the employee and agent;
• the prompt source;
• when it occurred;
• which action Gesta applied;
• what audit material is available.

Depending on the finding and organization configuration, audit material may include an original prompt sample or metadata and a stable fingerprint. Limit access to findings and treat exported evidence as sensitive.

Employee experience

When a prompt is blocked, the employee should:

1. read the matched rule and category;
2. remove or replace the sensitive value;
3. use an approved secure reference when available;
4. submit the sanitized prompt again;
5. contact the rule owner if the match appears incorrect.

A blocked prompt should not be bypassed by changing formatting or splitting the same sensitive value across multiple messages.

Sensitive Data and other controls

• Use Organization Library to explain how employees and agents should handle data.
• Use Sensitive Data to apply a prompt-level Record, Warn, or Block outcome.
• Use Operational Policies for commands that access, copy, or modify protected systems.

See Security and privacy for access, evidence, and retention recommendations.