CrowdStrike Falcon AIDR

Block prompt injection and PII before it reaches your model

Some setup needed Web · API

workflow #ai-security#prompt-injection-defense#agent-monitoring

About

Insert AI Guard between your app and LLM calls to redact PII, enforce content rules, and block prompt injection in real time. AI developers and security teams use it to apply configurable “recipes,” log every detection, and trigger webhooks for alerts across chatbots and agent apps. Portkey integration and an audit trail make it easy to add protection without rewriting your stack.

Editor's Take

We recommend AI Guard for teams that need enforceable, auditable protections across chatbots and agent apps without rebuilding downstream models; best suited when you can add an API proxy and want webhook alerts plus an audit trail.

Key Features

Route user prompts and model responses through the API → prompt injection is blocked and PII is redacted per configured recipes
Enable Confidential/PII rules (email, location, phone) → sensitive fields are replaced before content reaches your model
Turn on Malicious Entity detector for outputs → IPs are defanged and harmful payloads are flagged in real time
Add webhooks on detection events → your incident channel receives instant alerts with context
All detections auto-log to an audit trail → retrieve evidence for compliance reviews and debugging in minutes

Use Cases

A platform engineer securing a customer-support chatbot against prompt injection across web and mobile clients
A data privacy officer enforcing automatic PII redaction in user prompts before sending them to LLM providers
A security analyst receiving webhook alerts and reviewing audit logs when malicious content appears in model outputs

Try It Like This

1
Protect a support chatbot
Role: Platform engineer automating protection for a customer-support chatbot across web and mobile → Add AI Guard as a proxy for all LLM requests so incoming prompts are scanned and PII is redacted before reaching the model → Enable webhook alerts and audit logging to notify security channels and keep evidence for post-incident review.
2
Automate PII redaction pipeline
Role: Data privacy officer automating prompt sanitization between frontend and LLM provider → Route user inputs through AI Guard with Confidential/PII recipes turned on to replace emails, phone numbers, and locations → Verify redaction with audit logs and replay evidence during compliance checks.
3
Block prompt injection for agent apps
Role: Security engineer securing autonomous agents communicating with LLMs → Insert AI Guard between the agent and the model to detect and block prompt-injection attempts in real time → Configure the Malicious Entity detector to flag harmful outputs and forward incidents to a webhook for immediate investigation.
4
Alert on harmful model outputs
Role: Security analyst running monitoring for hosted LLM outputs → Send model responses through AI Guard so the Malicious Entity detector can defang IPs and flag harmful payloads → Trigger webhooks for the incident channel and review the correlated audit trail for context.
5
Compliance-ready audit trail
Role: Compliance manager building evidence pipelines for audits → Use AI Guard to log every detection and redaction event into the audit trail → Pull logs and supporting evidence when preparing compliance reports or responding to data subject requests.

Pros & Cons

Pros

API-based insertion between app and LLM calls enables real-time prompt-injection blocking without rewriting the model integration
Configurable recipes include Confidential/PII rules (email, location, phone) so sensitive fields are redacted before reaching the model
Detections auto-log to an audit trail and can trigger webhooks, providing immediate alerts plus retrievable evidence for compliance and debugging

Cons

Requires integrating AI Guard into the request path (API integration), which adds development work to existing LLM flows

Getting Started

1 Create a free Pangea account, open the console, and enable AI Guard
2 Copy the domain and service token, then call the AI Guard SDK/API and enable the pangea_prompt_guard and pangea_llm_response_guard recipes
3 Send a test message with an email or injection string and see it redacted/blocked with a logged detection (and optional webhook alert)