CrowdStrike AIDR integration

View as Markdown

The CrowdStrike AIDR guardrail uses configurable detection policies to identify and mitigate risks in AI application traffic, including:

  • Prompt injection and jailbreak attempts - Adversarial prompts designed to manipulate AI behavior or bypass security controls
  • Sensitive data exposure - PII, credentials, financial data, and confidential information in prompts and responses via built-in patterns, natural language processing, and custom definitions
  • Malicious entities - Known malicious URLs, IP addresses, and domains in AI outputs using integrated threat intelligence
  • Toxic and harmful content - Violent, abusive, or harmful content in AI inputs and outputs
  • Language - Language detection with optional use of an allowlist or denylist
  • Topic violations - Configurable content category restrictions

All detections are logged in an audit trail for analysis, attribution, and incident response. Note that this guardrail operates in a fail-open mode.

The following environment variable is required to use the CrowdStrike AIDR integration:

  • CS_AIDR_TOKEN: CrowdStrike AIDR API token.

You can also optionally set:

  • CS_AIDR_BASE_URL_TEMPLATE: Template for constructing the base URL for API requests. The {SERVICE_NAME} placeholder will be replaced with the service name slug. Defaults to https://api.crowdstrike.com/aidr/{SERVICE_NAME}.

Setup

Colang v1:

1# config.yml
2
3rails:
4 config:
5 crowdstrike_aidr:
6 timeout: 30.0 # Optional request timeout in seconds. Defaults to 30 seconds.
7
8 input:
9 flows:
10 - crowdstrike aidr guard input
11
12 output:
13 flows:
14 - crowdstrike aidr guard output

Colang v2:

1# config.yml
2
3colang_version: "2.x"
4
5rails:
6 config:
7 crowdstrike_aidr:
8 timeout: 30.0 # Optional request timeout in seconds. Defaults to 30 seconds.
# rails.co
import guardrails
import nemoguardrails.library.crowdstrike_aidr
flow input rails $input_text
crowdstrike aidr guard input
flow output rails $output_text
crowdstrike aidr guard output