Threat Alerts

Here you'll learn about Javelin's Threat Alerts, which give you real-time visibility and notifications so you can quickly react to security incidents and keep a strong AI security profile.

Threat alerts are the main way you can get an understanding of the issues your AI resources are facing, how often they're happening, and where they're coming from. When a Guardrail detects a policy violation, Javelin creates a detailed threat alert in addition to blocking or modifying the request.

Centralized Threat Dashboard

Your threat alerts are gathered in a centralized dashboard within Javelin. From there you can get a comprehensive view of your organization's threat status and take actions like:

• Viewing a real-time feed of alerts so you can see security events as they happen

• Filtering and searching so you can easily find alerts by threat type, application, route, timeframe, or other criteria

• Analyzing trends so you can spot patterns in attack types or resources targeted and strengthen your defenses where it's needed most

• Drilling down for details so you can understand the full context of requests, including the prompt or response that caused the issue, the guardrail that was triggered, and the action that occurred

Guardrail Failure Intelligence

In addition to detecting threats from user inputs, Javelin gives you vital insights about your organization's security health. A Guardrail failure happens when a security policy can't be run correctly due to a processor misconfiguration, internal processing error network issue, missing dependency, or endpoint failure.

These failures represent potential unseen risks in your security. Javelin automatically looks for Guardrail failures and organizes them in a dedicated view within Threat Alerts > Requests with Guardrail Failure so that your security and platform teams can:

• Quickly identify and triage misconfigured guardrails, with details about which policy is failing and why, plus error codes and messages

• Keep track of gaps in policy enforcement, by understanding where your security may need attention due to technical issues

• Remediate issues, so you can take proactive steps before they can lead to a security incident

Alert Categories

• Sensitive Data (Reject, Masked, Replaced, Redacted)

• Restricted Keywords

• Prompt Injections

• Jailbreak Attempts

• Sexual Content

• Profanity

• Violence

• Hate Speech

• Crime

• Weapons

• Markdown/Code

• Non-Ascii Character

• Invisible Characters

• Phishing URLs

• Non-English Language

• High Entropy

• Custom Guardrails

• Requests With Guardrail Failure

• Command Injection

• SQL Injection

• Path Traversal

• Secrets Leakage

Proactive Alerting and Notifications

You can set up Javelin to send real-time notifications to Slack and Splunk when it spots specific threats.

What's Next?

• Learn about the Guardrails & Policies that generate these alerts.

• Get the full context of an alert in a request trace by reading about Observability.