AI Telemetry Usage
For the LA release of Gigamon Insights, you must enable AI Usage Telemetry. This will periodically share non-sensitive AI usage data with Gigamon for the purpose of improving AI quality. Gigamon is committed to security and privacy. Any Personally Identifiable Information (PII) is sanitized and no data is shared with third-parties. All possible security measures are also implemented to protect and isolate the data collected.
Telemetry collection is designed to minimize data storage, exclude AI-generated responses, and protect user privacy by automatically removing personally identifiable information (PII).
The system collects the following categories of telemetry.
User Prompts (after PII Scrubbing)
The text you enter as a prompt is processed by an automated PII scrubbing pipeline, and any sensitive or identifiable information is removed before it is used. Only the sanitized version of the prompt is stored. PII scrubbing process detects and removes or masks commonly sensitive identifiers, including:
|
■
|
Government identifiers (SSN)
|
|
■
|
Street/mailing addresses |
Detected PII elements (for example, IP addresses or email addresses) are replaced with generic placeholders such as <IP>, <MAC_ADDRESS>, or <EMAIL_ADDRESS>.
For example,
Original Prompt: Search all available metadata logs for any records associated with the specified Email address, IP address or MAC address, including historical and current entries. Search Criteria:
|
■
|
MAC Address: 20:e5:2a:b6:93:f1 |
|
■
|
Email Address: john.doe@acme.com |
Stored Telemetry Version: Search all available metadata logs for any records associated with the specified Email address, IP address or MAC address, including historical and current entries. Search Criteria:
|
■
|
MAC Address: <MAC_ADDRESS> |
|
■
|
Email Address: <EMAIL_ADDRESS> |
Operational Metadata
Operational metadata describes how each AI request is processed and includes:
|
■
|
Message ID – Unique identifier for the AI request. |
|
■
|
Conversation ID – Identifier for the chat session. |
|
■
|
Parent Message ID – Identifier for the preceding message. |
|
■
|
Tenant ID/Tenant code – Unique deployment identifiers assigned during setup. |
|
■
|
Timestamps – When the message was created and last updated. |
|
■
|
Provider identifier – AI or LLM provider (for example, Bedrock). |
|
■
|
Model identifier – Model used for inference.
|
|
■
|
Token counts – Number of input (prompt), output (completion), and total tokens.
|
Operational metadata and the scrubbed prompt text are stored as structured JSON records.
User Feedback
If you choose to rate or comment on an AI response, the following optional telemetry may be collected:
|
■
|
User rating – Thumbs up or thumbs down. |
|
■
|
Optional comment – Free text feedback about the response. |
Feedback comments also undergo PII scrubbing before storage. Feedback telemetry is used to identify areas where AI responses can be improved.
