Skip to main content

Overview

The Threat Alert Page provides a comprehensive overview of security threats identified during interactions with Language Models. It focuses on detecting and analyzing vulnerabilities such as prompt injections, jailbreak attempts, code/markdown content, restricted keywords, and exposures of sensitive data, including PII, PHI, and regex-based patterns. This page enables users to efficiently evaluate risks across different threat categories and take necessary actions to safeguard their systems and sensitive information.

Key Features​

Centralized Hub for Security Incidents​

The Threat Alerts feature is your new command center for monitoring and managing sensitive requests and security incidents.

Threat Alert Overview Image

Centralized Tracking of Sensitive Requests​

Aggregate key security events into one easy-to-navigate page.

Alert Categories​

  • Sensitive Data
  • Prompt Injections
  • Jailbreak Attempts
  • Restricted Keywords
  • Markdown/Code Alerts

Interactive Filters​

Click on any category metric card to filter the requests that match the metric accordingly.

Comprehensive Request List View​

We provide a detailed list view, including fields with quick access to easily scan and identify requests that require attention:

  • Request Time
  • Model Name (e.g., gpt4o)
  • Provider (e.g., OpenAI)
  • Request Status (success, exceptions)
  • Sensitive Data Indicators (masked, replaced, redacted)
  • Duration

Threat Alert Request List Image

Detailed Request Insights​

You can click on any row to open a Right Panel View that displays detailed information including request, response data, processor outputs to see which guardrails were applied with name, latency and success/failure. Additional metadata is also available to identify:

  • Infotype Categories detected (PII, PHI etc.,)
  • Reserved Keywords or Regex Matches Each request is automatically tagged with relevant categories for easy identification, which helps you focus on the most critical issues first.

Threat Alert Processor Output Image

Flexible Time Filters​

Customize your view with time-based filtering for last 24 hours, 7 days, 30 days or 365 days.

Threat Alert Time Filter Image