GitHub - aiauthz/llm-authz-audit: Scan your LLM-powered applications for authorization gaps, leaked credentials, missing rate limits, prompt injection risks, and other security issues — before they reach production.

  ╦   ╦   ╔╦╗   ╔═╗ ╦ ╦ ╔╦╗ ╦ ╦ ╔═╗   ╔═╗ ╦ ╦ ╦═╗  ╦  ╔╦╗
  ║   ║   ║║║   ╠═╣ ║ ║  ║  ╠═╣ ╔═╝   ╠═╣ ║ ║ ║ ║  ║   ║
  ╩═╝ ╩═╝ ╩ ╩   ╩ ╩ ╚═╝  ╩  ╩ ╩ ╚══   ╩ ╩ ╚═╝ ╩═╝  ╩   ╩

# With npx (no install needed)
npx llm-authz-audit scan .

# Or install via pip
pip install llm-authz-audit
llm-authz-audit scan .

╭──────────────────╮
│ 16 Code Findings │
╰──────────────────╯

    api/__init__.py
   ❯❯❱ EP001  [LLM06]
          Unauthenticated LLM endpoint
          29┆ @app.route('/api/v1/predict', methods=['POST'])
          fix: Add authentication dependency: Depends(get_current_user)

    api/model_service.py
   ❯❱ AL001  [LLM09]
          LLM API call without logging
          16┆ r = openai.Moderation.create(
          fix: Add logging around LLM API calls for audit purposes.

╭──────────────╮
│ Scan Summary │
╰──────────────╯
  ⚠ Findings: 16 (2 blocking)
  • Analyzers run: 8
  • Files scanned: 13
  • ❯❯❱ High: 2
  • ❯❱ Medium: 14

llm-authz-audit scan . --format json

llm-authz-audit scan . --format sarif > results.sarif

npx llm-authz-audit scan .

# Install globally
pip install llm-authz-audit

# Or use pipx for isolation
pipx install llm-authz-audit

git clone https://github.com/aiauthz/llm-authz-audit.git
cd llm-authz-audit
pip install -e ".[dev]"

llm-authz-audit scan [PATH] [OPTIONS]

# Scan current directory
llm-authz-audit scan .

# Scan with SARIF output, fail only on critical
llm-authz-audit scan ./my-app --format sarif --fail-on critical

# Scan with specific analyzers
llm-authz-audit scan . --analyzers SecretsAnalyzer,EndpointAnalyzer

# Exclude test files
llm-authz-audit scan . --exclude "tests/*,*.test.py"

# Filter out low-confidence noise
llm-authz-audit scan . --min-confidence medium

# Only scan files changed since main
llm-authz-audit scan . --diff main

# Suppress known findings
llm-authz-audit scan . --suppress .llm-audit-suppress.yaml

# Load custom rules
llm-authz-audit scan . --extra-rules ./my-rules,./team-rules

llm-authz-audit list-analyzers

llm-authz-audit list-rules

# Include custom rules
llm-authz-audit list-rules --extra-rules ./my-rules

api_key = "sk-proj-abc123..."  # nosec — used for testing only

# .llm-audit-suppress.yaml
suppressions:
  - rule_id: SEC001
    file_pattern: "tests/*"
    reason: "Test fixtures with fake keys"

  - rule_id: EP001
    reason: "Public API — auth handled by API gateway"

  - file_pattern: "scripts/*"
    reason: "Internal tooling, not deployed"

llm-authz-audit scan . --suppress .llm-audit-suppress.yaml

# Environment variable — not flagged
api_key = os.environ["OPENAI_API_KEY"]

# Auth decorator — EP001 suppressed
@app.post("/chat")
@login_required
def chat_endpoint(request): ...

# Rate limiter present — RL001 suppressed
@limiter.limit("10/minute")
@app.post("/chat")
def chat_endpoint(request): ...

# Output format: console, json, or sarif
format: console

# Minimum severity to cause non-zero exit
fail_on: high

# Analyzers to enable (omit to enable all)
# analyzers:
#   - SecretsAnalyzer
#   - EndpointAnalyzer
#   - PromptInjectionAnalyzer

# Glob patterns to exclude
exclude:
  - "tests/*"
  - "*.test.py"

# AI-powered deep analysis
ai:
  enabled: false
  provider: anthropic
  model: claude-sonnet-4-5-20250929

# Using Anthropic (default)
export ANTHROPIC_API_KEY=your-key
llm-authz-audit scan . --ai

# Using OpenAI
export OPENAI_API_KEY=your-key
llm-authz-audit scan . --ai --ai-provider openai

# Limit AI cost (default: 20 findings max)
llm-authz-audit scan . --ai --ai-max-findings 10

pip install llm-authz-audit[ai]

name: LLM Security Audit
on: [push, pull_request]

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - run: pip install llm-authz-audit
      - run: llm-authz-audit scan . --format json --fail-on high

name: LLM Security Audit
on: [push, pull_request]

jobs:
  audit:
    runs-on: ubuntu-latest
    permissions:
      security-events: write
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v5
        with:
          python-version: "3.12"
      - run: pip install llm-authz-audit
      - run: llm-authz-audit scan . --format sarif > results.sarif
        continue-on-error: true
      - uses: github/codeql-action/upload-sarif@v3
        with:
          sarif_file: results.sarif

- run: llm-authz-audit scan . --diff origin/main --format sarif > results.sarif
  continue-on-error: true

# my-rules/custom.yaml
rules:
  - id: CUSTOM001
    title: "Internal API called without auth header"
    severity: high
    owasp_llm: LLM06
    file_types: ["*.py"]
    pattern: "requests\\.(?:get|post)\\(.+internal-api"
    negative_pattern: "headers.*[Aa]uth"
    remediation: "Add Authorization header to internal API calls."

llm-authz-audit scan . --extra-rules ./my-rules
llm-authz-audit list-rules --extra-rules ./my-rules