Laravel AI Lint

Static analysis for insecure AI wiring in a Laravel app. Two halves:

1. A scan command that finds AI provider keys leaked into tracked or browser-reachable files, plus other secrets, and fails CI.
2. A PHPStan extension that flags model output flowing into an unsafe sink and prompts built by string concatenation.

It is the static companion to laravel-ai-egress-guard: the egress guard catches leaks at runtime, AI Lint catches the dangerous wiring in source before it ships.

This is an independent open-source package. It is not affiliated with, endorsed by, or sponsored by Laravel, Laravel LLC, or any AI provider.

Requirements

• PHP 8.2+, Laravel 12 or 13
• PHPStan 2 (for the extension half)

Installation

composer require --dev laravelsecurityaudit/laravel-ai-lint

It depends on laravelsecurityaudit/laravel-secret-scanner for the detection engine. With phpstan/extension-installer present, the PHPStan rules register automatically; otherwise include the extension manually (see below).

The scan command (the funnel half)

php artisan ai-lint:scan
php artisan ai-lint:scan --min-severity=critical --format=sarif --output=ai-lint.sarif

It walks the configured paths (app, config, routes, resources, database, public by default), runs the secret rules over each file, and reports findings with their file and line. secrets.ai_provider_key fires on an OpenAI, Anthropic, or Google key found in a tracked file. When it lands under resources/js, resources/views, or public, that is the day-one mistake in AI-generated apps: the key is reachable by the browser.

Formats are table, json, and sarif. The command exits non-zero when any finding meets --min-severity, so it gates CI:

- run: php artisan ai-lint:scan --min-severity=critical --format=sarif --output=ai-lint.sarif
- uses: github/codeql-action/upload-sarif@v3
  with:
    sarif_file: ai-lint.sarif

The PHPStan rules (the static half)

Once installed (with phpstan/extension-installer, or by adding vendor/laravelsecurityaudit/laravel-ai-lint/extension.neon to your phpstan.neon includes), running PHPStan flags:

• aiLint.llmOutputToUnsafeSink: model output used directly inside an unsafe sink, e.g. DB::raw($response->content) or exec($prism->asText()->text) (OWASP LLM05).
• aiLint.concatenatedPrompt: a prompt built by concatenation or interpolation that mixes in dynamic data, e.g. ->withPrompt('Summarise: '.$input) (OWASP LLM01).

The detected method, sink, and token lists live in the package's extension.neon service definitions; copy a definition into your own phpstan.neon to customise them.

These are lexical, single-expression checks: they catch the AI call inside the sink or prompt argument. Output that first travels through a variable, and deeper data flow, is exactly what the runtime egress guard and a manual audit are for. AI Lint is the cheap first gate, not a replacement for either.

Configuration

php artisan vendor:publish --tag=ai-lint-config

Set scanned paths, file extensions, client-reachable locations, and the active rules in config/ai-lint.php.

Testing

composer test
composer analyse

The Laravel Security Audit family

One detection engine, guarding every place data leaves your app.

License

The MIT License (MIT). See LICENSE.