AI Guardrails & Content Governance

Protect your applications from prompt injection, sensitive data leakage (PII/DLP), and toxic content using Naagmani's multi-layered guardrail pipeline.

• Portal Page: /projects/[projectId]/governance

graph LR
    Prompt["User Prompt"] --> PII["PII / DLP Anonymizer (SSN, Credit Card)"]
    PII --> PromptShield["Prompt Injection Defense"]
    PromptShield --> Model["LLM Inference Engine"]
    Model --> Toxicity["Output Toxicity & Compliance Filter"]
    Toxicity --> Response["Safe Response to User"]

Active Guardrail Filters #

1. PII Masking & Redaction: Automatically redacts social security numbers, credit card details, phone numbers, and email addresses before sending prompts upstream.
2. Prompt Injection & Jailbreak Defense: Detects adversarial jailbreak techniques and overrides.
3. Output Content Moderation: Scans model responses for hate speech, violence, and proprietary corporate secrets.

Next Steps #

• Build model routing rules: Smart Routing Policies
• Review audit logs: Security Audit Logs