Guardrails & Policies

Here you'll learn how Javelin helps you detect threats and defend your resources by setting the rules of the road for every interaction in your AI traffic.

Guardrails are Javelin's main tool for governing your AI security strategy. They work like a series of automated checkpoints — as traffic flows through the Gateway, it passes through these checkpoints, with each inspecting the request or response for a specific issue. Key functions of Guardrails include:

• Threat Protection: Scan for and block malicious code

• Data Loss Prevention (DLP): Identify and redact sensitive information before it leaves your system

• Content Moderation: Filter requests and responses to ensure they align with your organization's policies

• Input Validation: Set rules to ensure that your models don't accept malicious or anomalous inputs

• Custom Policy Enforcement: Create custom rules to enforce requirements specific to your organization

Detect Threats in Real Time

Defense begins with detection. Javelin gives you a wide range of built-in Guardrails that are made to identify common AI security threats before they can impact your enterprise.

Some of these detection features include:

• Prompt Injection and Jailbreak Detection: Identify and block attempts to manipulate models

• Malware and Phishing Detection: Scan for links to known malicious domains or social engineering sites

• Security Filters: Spot unsafe code snippets and invisible characters

Defend Your Data and Brand

When a threat is detected, or as part of proactively protecting your data, Javelin's guardrails get the job done. You can set them up to block, modify, or redact content to ensure safety and compliance.

Some of these defense capabilities include:

• Sensitive Data Protection: Detect and safeguard information like PII, PHI, and other secrets

• Trust & Safety Filters: Check for harmful, toxic, and otherwise inappropriate content

• Keyword & Regex Filters: Set custom lists of words or patterns to block or flag

• Language Detection: Ensure your applications are using expected languages, and prevent misuse of unsupported languages

Standalone Guardrails

Javelin also lets you check content for security guidelines on their own, without having to go through the usual LLM request and response process. You can screen content, validate policies, and proactively detect threats without having to make full REST API calls for Guardrails.

By combining these detection and defense capabilities, you can create a robust security posture that protects your AI traffic from threats, ensures data privacy, and enforces compliance with your policies.

What's Next?

• Learn on the AI Gateway page how to apply these policies to your traffic.

• Read about Threat Alerts to learn what happens when a guardrail is triggered.