Secure a Laravel Application: OWASP Risks and Practical Defences
Laravel gives you secure primitives; it cannot decide who may access an invoice, which URL a server may fetch, or what a browser is allowed to load.
Authorize on the server
Use policies for model actions and gates for capabilities. The controller invokes authorization before it reads or mutates data.
<?php
declare(strict_types=1);
namespace App\Policies;
use App\Models\User;
final class UserPolicy
{
public function viewAny(User $user): bool
{
return $user->hasPermissionTo('users.view');
}
}
For a super-admin escape hatch, put the rule in one reviewable place:
Gate::before(fn (User $user): ?bool => $user->isSuperAdmin() ? true : null);
On Laravel 11+, controller middleware is declared with HasMiddleware; constructor $this->middleware() is not the current pattern.
final class AdminUserController extends Controller implements HasMiddleware
{
public static function middleware(): array
{
return ['auth', 'can:viewAny,'.User::class];
}
}
Client-side hiding is UX, never authorization. Feature-test a denied user for every sensitive route.
Validate, bind and encode
Form Requests are the HTTP boundary. Use $request->validated(), not $request->only(), so unvalidated fields cannot slip into mass assignment. Eloquent and the query builder parameterize values; never concatenate input into SQL, shell commands or dynamic column names. Escape output by default and sanitize deliberately if user HTML is a product feature.
Credentials and sessions
Passwords use Laravel’s hashed cast or Hash::make(), and hashes are not recoverable. API token systems should store token secrets hashed at rest; show the plaintext token only once. Encrypt reversible integration credentials with an encrypted cast, protect APP_KEY, and rotate secrets with an overlap plan.
Use Sanctum for first-party SPAs and personal access tokens when its model fits. Give tokens narrow abilities, revoke them on compromise, rate-limit login and reset endpoints, and enable email verification/MFA where the risk warrants it. Never log Authorization headers, cookies or reset URLs.
Browser and deployment boundaries
Serve only public/, set APP_DEBUG=false, use HTTPS and secure cookie settings. Add security headers deliberately: a Content-Security-Policy tailored to actual scripts, X-Content-Type-Options: nosniff, clickjacking protection, a strict referrer policy and HSTS after HTTPS is stable. Start CSP in report-only mode, inspect violations, then enforce it; copying a permissive policy defeats its purpose.
Audit Composer and npm dependencies in CI, pin intentional upgrades, and remove unused packages. composer audit reports known advisories; it is not a complete application security review.
SSRF, uploads and external input
An endpoint that fetches a user-provided URL can reach internal infrastructure. Parse and allow-list intended hosts, resolve DNS defensively, reject loopback, link-local and private address ranges, disable redirect surprises, set short timeouts and restrict egress at the network layer. A hostname regex alone is not protection.
Validate upload MIME type and size, store uploads outside the public root, generate server-side names and scan high-risk files. Signed URLs are useful for time-limited access, not a substitute for checking authorization.
Observe and rehearse
Log failed authentication and authorization with minimal identifiers; report unexpected exceptions to your error tracker without secrets or personal payloads. Queue listeners and webhooks must be idempotent. Security work is complete only when a test proves the bad path is rejected and an operator can investigate it safely.
Production checklist
- Policies/Gates cover every sensitive action and are feature-tested.
- Requests use validated data; database and shell input is never concatenated.
- Tokens are scoped and hashed; secrets are encrypted or environment-owned.
- HTTPS, debug settings, public web root and headers are reviewed per environment.
- SSRF, upload and webhook boundaries have explicit tests and operational limits.