no-sensitive-in-prompt

📡 Live from GitHub — This documentation is fetched directly from no-sensitive-in-prompt.md and cached for 6 hours.

Prevents sensitive data (passwords, tokens, PII) from being sent to LLMs.

This rule identifies code patterns where sensitive data like passwords, API keys, tokens, or personally identifiable ...

📊 Rule Details

Property	Value
Type	problem
Severity	🔴 CRITICAL
OWASP LLM	LLM02: Sensitive Information Disclosure
CWE	CWE-200: Information Exposure
CVSS	8.5
Config Default	`error` (recommended, strict)

This rule identifies code patterns where sensitive data like passwords, API keys, tokens, or personally identifiable information (PII) is passed to AI prompts. LLM providers may log, store, or use this data for training.

❌ Incorrect Code

// Password in prompt
await generateText({
  prompt: `Reset password for user. Current password: ${userPassword}`,
});

// API key in prompt
await generateText({
  prompt: `Configure service with key: ${apiKey}`,
});

// SSN in prompt
await streamText({
  prompt: `Process application for SSN: ${socialSecurityNumber}`,
});

// Credit card in prompt
await generateText({
  prompt: `Validate credit card: ${creditCardNumber}`,
});

✅ Correct Code

// Redacted data
await generateText({
  prompt: `Reset password for user. Current password: [REDACTED]`,
});

// No sensitive data
await generateText({
  prompt: `Configure service with the API key stored in environment variables.`,
});

// Use redaction helper
await streamText({
  prompt: `Process application for SSN: ${redact(socialSecurityNumber)}`,
});

// Reference instead of value
await generateText({
  prompt: `Validate the credit card on file for user ${userId}`,
});

⚙️ Options

Option	Type	Default	Description
`sensitivePatterns`	`string[]`	`['password', 'secret', 'token', 'apiKey', 'api_key', 'credential', 'ssn', 'socialSecurity', 'creditCard', 'cvv', 'privateKey']`	Variable patterns suggesting sensitive data

🛡️ Why This Matters

Sending sensitive data to LLMs can result in:

Data breach - LLM providers may store prompts
Training data poisoning - Your data may be used to train models
Compliance violations - GDPR, HIPAA, PCI-DSS violations
Third-party exposure - Data shared with third-party AI providers

no-hardcoded-api-keys - Prevent hardcoded credentials
require-output-filtering - Filter sensitive tool output

Known False Negatives

The following patterns are not detected due to static analysis limitations:

Variable Names Not Matching Patterns

Why: Only configured sensitive patterns are checked.

// ❌ NOT DETECTED - Custom field name
await generateText({
  prompt: `Process data: ${mySecretField}`, // Not in sensitivePatterns
});

Mitigation: Configure sensitivePatterns with custom field names.

Dynamic Prompt Construction

Why: Prompts built at runtime are not analyzed.

// ❌ NOT DETECTED - Dynamic prompt
const fields = [userPassword, apiKey];
const prompt = fields.join(', ');
await generateText({ prompt });

Mitigation: Never concatenate sensitive data into prompts.

Nested Object Properties

Why: Deep property access may not be recognized.

// ❌ NOT DETECTED - Nested property
await generateText({
  prompt: `Reset with: ${user.credentials.password}`,
});

Mitigation: Configure patterns to match nested sensitive fields.

Encrypted/Transformed Data

Why: Transformed data appears safe but may be sensitive.

// ❌ NOT DETECTED - Encrypted but still sensitive
await generateText({
  prompt: `Decrypt this: ${encryptedPassword}`,
});

Mitigation: Never send any form of credentials to LLMs.