Devin
Security FAQ

What are common security mistakes in Devin AI apps?

Get instant answers about your app's security.

Short Answer

The mistakes we see repeatedly in Devin AI apps: no human-in-the-loop for security decisions; exposed api endpoints; insecure dependency choices. Each one is a specific failure mode of Devin AI's workflow — not generic programming mistakes.

Detailed Answer

The mistakes we actually see in Devin AI apps

These aren't hypothetical — they're what VAS finds when it scans a Devin AI app for the first time. Listed in order of how often they appear:

1. No Human-in-the-Loop for Security Decisions

*Why it happens:* Devin makes architectural choices autonomously that have security implications.

*Fix:* Verify with a scan — catching this manually requires knowing it exists, which is the problem.

2. Exposed API Endpoints

*Why it happens:* Autonomously created API routes may lack auth middleware.

*Fix:* Verify with a scan — catching this manually requires knowing it exists, which is the problem.

3. Insecure Dependency Choices

*Why it happens:* Devin may install packages with known vulnerabilities.

*Fix:* Verify with a scan — catching this manually requires knowing it exists, which is the problem.

4. Missing Access Controls on Data

*Why it happens:* Database tables and APIs may be accessible without authorization.

*Fix:* Verify with a scan — catching this manually requires knowing it exists, which is the problem.

Why these specifically show up in Devin AI (and not as much elsewhere)

Devin AI's workflow optimizes for speed — idea to deployed app in minutes. The mistakes above aren't character flaws, they're the predictable output of a speed-optimized workflow that doesn't enforce security gates. The fix is treating security gates as non-negotiable, not as "I'll get to it later."

Security Research & Statistics

10.3%

of Lovable applications (170 out of 1,645) had exposed user data in the CVE-2025-48757 incident

Source: CVE-2025-48757 security advisory

4.45 million USD

average cost of a data breach in 2023

Source: IBM Cost of a Data Breach Report 2023

500,000+

developers using vibe coding platforms like Lovable, Bolt, and Replit

Source: Combined platform statistics 2024-2025

Expert Perspectives

There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.

Andrej KarpathyFormer Tesla AI Director, OpenAI Co-founder

Vibe coding your way to a production codebase is clearly risky. Most of the work we do as software engineers involves evolving existing systems, where the quality and understandability of the underlying code is crucial.

Simon WillisonSecurity Researcher, Django Co-creator

Check Your Devin AI App's Security

VAS scans for all the security issues mentioned above. Get a comprehensive security report in minutes.

Get Starter Scan

More Questions About This Topic

How common are these mistakes in Devin AI apps — is this overstated?

Understated, if anything. The majority of Devin AI apps scanned for the first time have at least one of the high-likelihood mistakes above. "No Human-in-the-Loop for Security Decisions" in particular is the default state of a new Devin AI app before any security work. Our sample skews toward apps whose owners care enough to scan — the base rate for never-scanned Devin AI apps is higher.

What are the actual consequences when these mistakes ship to production?

The consequence ladder: (a) data exposure — emails, passwords, PII, payment info readable by anyone; (b) account takeover — if auth is weak, legitimate accounts get hijacked; (c) third-party abuse — an exposed OpenAI or Stripe key gets drained of quota or money; (d) regulatory — GDPR/CCPA notification requirements trigger at ~first exposure; (e) reputational — "Devin AI app data breach" is a headline that doesn't age well. Each consequence compounds the next.

How do I avoid these mistakes when building with Devin AI?

Three non-negotiable habits: (1) Configure Row Level Security (RLS) policies at table/collection creation — before writing any feature code. (2) Treat any paste-a-key-into-code as a bug from the first keystroke, not "I'll move it to env vars later." (3) Run a VAS scan before every production deploy — five minutes of scanning prevents hours-to-weeks of breach response. Specifically: start with no human-in-the-loop for security decisions.