Secrets Detection Guardrail - TrueFoundry Docs

This guide explains how to use TrueFoundry’s built-in Secrets Detection guardrail to protect sensitive credentials from being exposed in LLM interactions and MCP tool invocations.

Secrets Detection can be applied to all four guardrail hooks: LLM Input, LLM Output, MCP Pre Tool, and MCP Post Tool—providing comprehensive protection across your entire AI workflow.

What is Secrets Detection?

Secrets Detection is a built-in TrueFoundry guardrail that identifies and handles sensitive credentials and secrets in text content. It runs directly within the AI Gateway without requiring external API calls, providing fast and cost-effective secret detection.

Key Features

Comprehensive Secret Recognition: Detects a wide range of sensitive credentials including:
- AWS access keys and secret keys
- API keys (OpenAI, Anthropic, Google, etc.)
- JWT tokens and bearer tokens
- Private keys (RSA, SSH, PGP)
- Database connection strings
- OAuth tokens and refresh tokens
- Generic high-entropy strings that may be secrets
Flexible Operation Modes:
- Validate: Detect secrets and block requests/responses that contain them
- Mutate: Detect secrets and redact them with configurable replacement text, allowing the request to continue
Zero External Dependencies: Runs entirely within the AI Gateway with no external API calls, ensuring low latency and no additional costs.

Adding Secrets Detection Guardrail

To add Secrets Detection to your TrueFoundry setup, follow these steps:

Navigate to Guardrails

Go to the AI Gateway dashboard and navigate to the Guardrails section.

Create or Select a Guardrails Group

Create a new guardrails group or select an existing one where you want to add the Secrets Detection guardrail.

Add Secrets Detection Integration

Click on Add Guardrail and select Secrets Detection from the TrueFoundry Guardrails section.

TrueFoundry guardrail selection interface showing Secrets Detection option — Select Secrets Detection from TrueFoundry Guardrails

Configure the Guardrail

Fill in the configuration form:

Name: Enter a unique name for this guardrail configuration (e.g., secrets-detection)
Description: Optional description for this guardrail (default: “Detects likely credentials/secrets (AWS keys, API keys, JWT tokens, private keys)”)
Operation: Choose the operation mode:
- validate: Detect and block requests containing secrets
- mutate: Detect and redact secrets, allowing the request to continue
Redaction Text (for mutate mode): Text to replace detected secrets (default: ***REDACTED***)

Save the Configuration

Click Save to add the guardrail to your group.

Configuration Options

Parameter	Description	Default
Name	Unique identifier	Required
Operation	`validate` (block) or `mutate` (redact)	`validate`
Priority	Execution order for mutate guardrails (lower runs first)	`1`
Enforcing Strategy	`enforce`, `enforce_but_ignore_on_error`, or `audit`	`enforce`
Redaction Text	Replacement text (mutate mode only)	`*REDACTED*`

See Guardrails Overview for details on Operation Modes and Enforcing Strategy.

How It Works

Validate Mode

When configured in validate mode, the guardrail scans the content for secrets and blocks the request if any are detected. Example:

Input: "Use this API key: sk-abc123xyz456 to authenticate"
Result: Request blocked with error message indicating secret detected

Mutate Mode

When configured in mutate mode, the guardrail scans the content for secrets and replaces them with the configured redaction text. Example:

Input: "Use this API key: sk-abc123xyz456 to authenticate"
Output: "Use this API key: ***REDACTED*** to authenticate"

Detected Secret Types

The Secrets Detection guardrail identifies a comprehensive set of secret patterns across multiple categories. Below is the complete list of detected secret types.

Cloud Provider Credentials

Secret Type	Pattern Description	Example Format
AWS Access Key ID	AWS access key identifiers	`AKIA` followed by 16 alphanumeric characters
AWS Secret Access Key	AWS secret keys with contextual keywords	40-character base64 string near `secret` or `access` keywords
Google API Key	Google Cloud API keys	`AIza` followed by 35 characters
Google OAuth Token	Google OAuth access tokens	`ya29.` followed by 20+ characters
Azure SAS Token	Azure Shared Access Signature tokens	URL parameters with `sv=` and `sig=`
Azure Storage Account Key	Azure storage account keys	`AccountKey=` followed by 88-character base64 string
Azure Storage Connection String	Full Azure storage connection strings	`DefaultEndpointsProtocol=...AccountKey=...`
Digital Ocean Personal Access Token	DO API tokens	`dop_v1_` followed by 64 hex characters
Databricks Personal Access Token	Databricks API tokens	`dapi` followed by 32 alphanumeric characters
Cloudant Authenticated URL	IBM Cloudant database URLs with credentials	Cloudant URLs containing embedded credentials
Firebase Cloud Messaging Server Key	FCM server keys	`AAAA` prefix with 140+ character token
SoftLayer API Token	IBM SoftLayer API tokens	64-character tokens in SoftLayer API URLs

AI/ML Provider API Keys

Secret Type	Pattern Description	Example Format
OpenAI API Key	OpenAI API secret keys	`sk-` prefix with `T3BlbkFJ` marker, or `sk-` with 20+ characters
Anthropic API Key	Anthropic Claude API keys	`sk-ant-api` followed by version and 90+ characters
Airtable Secret Token	Airtable personal access tokens	`pat` followed by 14 characters, dot, and 64 hex characters
Groq API Key	Groq Cloud API keys	`gsk_` followed by 52 alphanumeric characters

Code Repository & Package Manager Tokens

Secret Type	Pattern Description	Example Format
GitHub Fine-Grained PAT	GitHub fine-grained personal access tokens	`github_pat_` followed by 20-255 characters
GitHub Personal Token	GitHub classic personal access tokens	`ghp_`, `gho_`, `ghu_`, `ghs_`, or `ghr_` followed by 36 characters
GitLab Token	Various GitLab token types	`glpat-`, `gldt-`, `glft-`, `glsoat-`, `glrt-`, `GR1348941`, etc.
NPM Token	NPM authentication tokens	`npm_` followed by 30+ characters
NPM .npmrc Auth Token	Tokens in .npmrc configuration	`_authToken=` with npm token or UUID format
PyPI Token	Python Package Index tokens	`pypi-` followed by base64 identifier and 70+ characters
Artifactory Credential	JFrog Artifactory API keys	`AKC` followed by 10+ alphanumeric characters
Artifactory Encrypted Password	Encrypted Artifactory passwords	`AP` followed by hex digit and 8+ characters

Communication & Collaboration Platforms

Secret Type	Pattern Description	Example Format
Slack Token	Slack API tokens (various types)	`xoxb-`, `xoxa-`, `xoxp-`, `xoxr-`, `xoxs-` formats
Slack Bot Token	Slack bot authentication tokens	`xox[abpors]-` followed by numeric segments
Slack Webhook URL	Slack incoming webhook URLs	`https://hooks.slack.com/services/T.../B.../...`
Discord Bot Token	Discord bot authentication tokens	Three-part token with `M`, `N`, or `O` prefix
Discord Webhook URL	Discord webhook URLs	`https://discord.com/api/webhooks/...`
Telegram Bot Token	Telegram bot API tokens	8-10 digit bot ID followed by 35-character token

Payment & E-Commerce

Secret Type	Pattern Description	Example Format
Stripe Publishable Key	Stripe public API keys	`pk_live_` followed by 20+ characters
Stripe Secret Key	Stripe secret/restricted API keys	`sk_live_` or `rk_live_` followed by 24+ characters
Square Access Token	Square API access tokens	`EAAA` followed by 60 characters
Square OAuth Secret	Square OAuth client secrets	`sq0csp-` followed by 43 characters
Shopify Token	Shopify API tokens	`shpat_`, `shppa_`, or `shpss_` followed by 32 hex characters

Database Connection Strings

Secret Type	Pattern Description	Example Format
MongoDB Connection String	MongoDB connection URLs with credentials	`mongodb://user:password@host`
MySQL Connection String	MySQL connection URLs with credentials	`mysql://user:password@host`
PostgreSQL Connection String	PostgreSQL connection URLs with credentials	`postgresql://user:password@host`

Private Keys & Certificates

Secret Type	Pattern Description	Example Format
Private Key (Generic)	RSA, EC, DSA, or encrypted private keys	`-----BEGIN PRIVATE KEY-----` blocks
RSA/DSA/EC Private Key	Specific algorithm private keys	`-----BEGIN RSA PRIVATE KEY-----` blocks
OpenSSH Private Key	OpenSSH format private keys	`-----BEGIN OPENSSH PRIVATE KEY-----` blocks
PGP Private Key	PGP/GPG private key blocks	`-----BEGIN PGP PRIVATE KEY BLOCK-----`
SSH2 Encrypted Private Key	SSH2 encrypted private keys	`-----BEGIN SSH2 ENCRYPTED PRIVATE KEY-----`
PuTTY Private Key	PuTTY PPK format private keys	`PuTTY-User-Key-File-` header
Certificate	X.509 certificates	`-----BEGIN CERTIFICATE-----` blocks

Authentication Tokens & Credentials

Secret Type	Pattern Description	Example Format
JWT Token	JSON Web Tokens	`eyJ...` three-part base64 structure
Basic Auth Credentials	Credentials embedded in URLs	`://username:password@` in URLs
Password	Password assignments in code/config	`password=`, `pass=`, `pwd=` with quoted values
Generic Secret	Generic secret/token/key assignments	`secret=`, `token=`, `key=` with 20+ character values

Other Services

Secret Type	Pattern Description	Example Format
Twilio Account SID	Twilio account identifiers	`AC` followed by 32 hex characters
Twilio API Key SID	Twilio API key identifiers	`SK` followed by 32 hex characters
Twilio Auth Token	Twilio authentication tokens	32-character hex tokens with `auth_token` context
SendGrid API Key	SendGrid email API keys	`SG.` followed by 22 and 43 character segments
Mailchimp API Key	Mailchimp API keys	32 hex characters followed by `-us` and datacenter number
Mapbox Access Token	Mapbox API tokens	`pk.` or `sk.` followed by 30+ characters
Notion Integration Token	Notion API integration tokens	`secret_` followed by 32+ characters
Linear API Key	Linear project management API keys	`lin_api_` followed by 40+ characters
Dropbox Access Token	Dropbox API tokens	`sl.` followed by 120+ characters
Facebook Access Token	Facebook/Meta API tokens	`EAA` followed by 30+ characters

Contextual Pattern Detection

In addition to specific patterns, Secrets Detection also identifies secrets based on contextual clues:

Pattern Type	Description
Hex Secret (32+ chars)	Hexadecimal strings near keywords like `secret`, `api_key`, `token`, `password`, `credential`, `auth_token`, `private_key`, `signing_key`, `encryption_key`
Base64 Secret (24+ chars)	Base64-encoded strings near the same contextual keywords

Contextual patterns help detect secrets that don’t follow a specific format but appear in contexts that suggest they are sensitive credentials.

Use Cases

Secrets Detection can be applied to any of the four guardrail hooks depending on your use case:

Recommended Hook Usage

Hook	Use Case
LLM Input	Prevent users from accidentally sending credentials to LLMs
LLM Output	Ensure LLM responses don’t leak sensitive information
MCP Pre Tool	Detect secrets in tool parameters before invocation
MCP Post Tool	Detect and redact secrets in tool outputs (file contents, API responses, etc.)

LLM Input Hook

Prevent users from accidentally sending sensitive credentials to LLMs:

curl -X POST "{GATEWAY_BASE_URL}/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H 'X-TFY-METADATA: {}' \
  -H 'X-TFY-GUARDRAILS: {"llm_input_guardrails":["my-guardrail-group/secrets-detection"],"llm_output_guardrails":[]}' \
  -H 'X-TFY-LOGGING-CONFIG: {"enabled": true}' \
  -d '{
    "model": "openai-main/gpt-4o-mini",
    "messages": [
      {"role": "system", "content": "You are an AI assistant."},
      {"role": "user", "content": "Help me debug this code that uses API key sk-abc123xyz"}
    ],
    "max_tokens": 2500,
    "stream": true
  }'

LLM Output Hook

Ensure LLM responses don’t leak sensitive information:

curl -X POST "{GATEWAY_BASE_URL}/chat/completions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H 'X-TFY-METADATA: {}' \
  -H 'X-TFY-GUARDRAILS: {"llm_input_guardrails":[],"llm_output_guardrails":["my-guardrail-group/secrets-detection"]}' \
  -H 'X-TFY-LOGGING-CONFIG: {"enabled": true}' \
  -d '{
    "model": "openai-main/gpt-4o-mini",
    "messages": [
      {"role": "system", "content": "You are an AI assistant."},
      {"role": "user", "content": "Generate an example AWS configuration file"}
    ],
    "max_tokens": 2500,
    "stream": true
  }'

MCP Tool Hooks

name: guardrails-control
type: gateway-guardrails-config
rules:
  - id: mcp-secrets-protection
    when:
      target:
        operator: or
        conditions:
          mcpServers:
            values:
              - file-reader
            condition: in
      subjects:
        operator: and
        conditions:
          in:
            - team:engineering
          not_in:
            - user:security-admin@example.com
    llm_input_guardrails: []
    llm_output_guardrails: []
    mcp_tool_pre_invoke_guardrails: []
    mcp_tool_post_invoke_guardrails:
      - my-guardrail-group/secrets-detection

For maximum protection, apply Secrets Detection to all four hooks to catch secrets from user input, LLM output, and MCP tool results.

Best Practices

Use mutate mode to redact secrets while preserving request flow—useful for logging scenarios.

High-entropy strings (UUIDs, encoded data) may trigger false positives. Test with your specific data.

​What is Secrets Detection?

​Key Features

​Adding Secrets Detection Guardrail

​Configuration Options

​How It Works

​Validate Mode

​Mutate Mode

​Detected Secret Types

​Cloud Provider Credentials

​AI/ML Provider API Keys

​Code Repository & Package Manager Tokens

​Communication & Collaboration Platforms

​Payment & E-Commerce

​Database Connection Strings

​Private Keys & Certificates

​Authentication Tokens & Credentials

​Other Services

​Contextual Pattern Detection

​Use Cases

​Recommended Hook Usage

​LLM Input Hook

​LLM Output Hook

​MCP Tool Hooks

​Best Practices

What is Secrets Detection?

Key Features

Adding Secrets Detection Guardrail

Configuration Options

How It Works

Validate Mode

Mutate Mode

Detected Secret Types

Cloud Provider Credentials

AI/ML Provider API Keys

Code Repository & Package Manager Tokens

Communication & Collaboration Platforms

Payment & E-Commerce

Database Connection Strings

Private Keys & Certificates

Authentication Tokens & Credentials

Other Services

Contextual Pattern Detection

Use Cases

Recommended Hook Usage

LLM Input Hook

LLM Output Hook

MCP Tool Hooks

Best Practices