The Vibe Coding Security Checklist: 10 Critical Flaws in AI-Generated Software

AI coding tools like Cursor, Claude Code, Lovable, v0, and ChatGPT are trained to solve one specific objective: generate code that fulfills the user's prompt as quickly as possible.

They are not incentivized to ensure defense-in-depth, least privilege, or cryptographic hygiene.

When teams move fast and ship "vibe-coded" prototypes directly to production, they frequently introduce severe vulnerabilities that automated security scanners and ethical hackers exploit in minutes.

Here is the 10-Point Security Audit Checklist we use at Silver Stack Software when auditing and hardening AI-generated software.

---

1. Exposed Client-Side Secrets & API Keys

The Vulnerability: AI assistants often place private API keys (Stripe secret keys, AWS credentials, database passwords) into client-side environment variables (NEXT_PUBLIC_* or REACT_APP_*) so frontend components can fetch data directly. The Fix: Audit all .env files. Ensure secrets are never prefixed with public flags, and route all third-party API calls through server-side API routes or Server Actions.

---

2. Insecure Direct Object References (IDOR)

The Vulnerability: AI frequently writes endpoints like GET /api/invoices/:id that fetch and return records using the requested ID without verifying that the authenticated user actually owns that invoice.

// ❌ VULNERABLE: Any logged-in user can view any invoice by guessing ID
export async function GET(req: Request, { params }: { params: { id: string } }) {
  const invoice = await db.invoice.findUnique({ where: { id: params.id } });
  return Response.json(invoice);
}

The Fix: Scope every database query by tenant and user identity:

// ✅ SECURE: Query scoped strictly to authenticated session
const invoice = await db.invoice.findFirst({
  where: { id: params.id, organizationId: session.user.orgId }
});
if (!invoice) return new Response('Not Found', { status: 404 });

---

3. Missing Server-Side Role-Based Access Control (RBAC)

The Vulnerability: AI models often hide admin buttons on the UI ({user.isAdmin && <AdminPanel />}), but forget to check user roles inside the underlying backend API endpoint. Malicious users can send raw POST requests directly to /api/admin/delete-user. The Fix: Enforce authorization middleware on every backend route, independent of frontend UI state.

---

4. Raw Unescaped SQL Queries & Injection Risks

The Vulnerability: When asked to implement complex filtering, AI models occasionally concatenate raw strings into SQL queries rather than using parameterized inputs.

// ❌ VULNERABLE: SQL Injection via string interpolation
const query = `SELECT * FROM users WHERE email = '${userEmail}'`;

The Fix: Always enforce parameterized queries or strictly typed ORM builders (Prisma, Drizzle, EF Core).

---

5. Unrestricted File Upload Endpoints

The Vulnerability: AI-generated file upload routes often trust client-provided file names and MIME types, allowing attackers to upload malicious .php, .sh, or executable files directly into public storage. The Fix: Generate random UUID file names on the server, validate file magic bytes (not just extensions), and store uploaded assets in isolated cloud buckets with public execution permissions disabled.

---

6. Weak Password Hashing & Insecure JWTs

The Vulnerability: AI models frequently implement authentication using outdated MD5 or SHA-256 hashing, or sign JSON Web Tokens with hardcoded fallback secrets like "secret123". The Fix: Use modern adaptive hashing algorithms (Argon2id or bcrypt with cost factor >= 12), and enforce cryptographically random 256-bit signing secrets.

---

7. Missing Rate Limiting & DoS Vulnerabilities

The Vulnerability: AI assistants write authentication and search endpoints without request throttling. Attackers can flood the endpoint with millions of requests, causing CPU exhaustion or massive third-party API bills. The Fix: Deploy edge rate-limiting middleware (via Redis / Upstash) restricting requests by IP and session.

---

8. Cross-Site Scripting (XSS) via dangerouslySetInnerHTML

The Vulnerability: When asked to render rich text or markdown, AI tools frequently reach for dangerouslySetInnerHTML without sanitizing input. The Fix: Use strict HTML sanitization libraries (DOMPurify) to strip executable JavaScript tags before rendering.

---

9. Dependency Supply Chain Hallucinations

The Vulnerability: Large Language Models sometimes hallucinate package names that do not exist in npm. Attackers register these hallucinated package names with malicious payloads. The Fix: Run npm audit and verify every package in package.json against official registries and download counts.

---

10. Missing Audit Logs & Unchecked Error Messages

The Vulnerability: AI code often logs full stack traces and internal database schemas directly into HTTP 500 error responses delivered to the client. The Fix: Catch all unhandled exceptions, log detailed traces to private server telemetry, and return generic, sanitized error messages to the browser.

---

The Verdict

Vibe coding accelerates prototype development from months to days. But before you launch to real customers, a rigorous security audit is non-negotiable.

Need an independent security and architecture audit of your AI-built software? Contact Silver Stack Software to schedule a comprehensive code review.