Set Sensitivity of a Data Type

Configure where data types should be treated as sensitive in agentic systems.

Control where data types are flagged as sensitive - in agent inputs, outputs, tool calls, or context.

Why Set Sensitivity

Example: Flag "TOKEN" as sensitive only in agent responses, not in requests. This reduces noise while ensuring leaked credentials are caught.

How to Configure

1. Go to Agentic Discovery > Sensitive Data

2. Click "..." on the data type

3. Select Edit

4. Choose sensitivity locations:

   • Agent Inputs (prompts, requests)

   • Agent Outputs (LLM responses)

   • Tool Invocations (MCP calls)

   • Context/Memory (conversation history)

   • All locations

Agentic-Specific Sensitivity

Agent Inputs: User prompts, questions

• Sensitive: PII, credentials, secrets

Agent Outputs: LLM-generated responses

• Sensitive: Leaked credentials, PII in responses

Tool Invocations: MCP tool calls

• Sensitive: Database queries, file paths, API keys

Context/Memory: Agent conversation history

• Sensitive: Accumulated PII, session data

Example: API Keys

Set API_KEY sensitive in:

• ✅ Agent Outputs (catch leaks)

• ✅ Tool Invocations (monitor access)

• ❌ Agent Inputs (expected to receive keys)