AI coding tools write fast.
GitHub Copilot. Cursor. ChatGPT. Claude. Gemini.
Every single one of them is now inside production
codebases at companies worldwide.
And every single one of them produces the same
security vulnerabilities repeatedly.
Not because they are bad tools.
Because they were trained on code that had
security mistakes in it — and they learned
those patterns too.
What Happened This Week
A developer handed me their production codebase
to audit.
I ran my new tool on it.
Here is what came back:
533 files scanned.
3 critical security vulnerabilities.
All produced by AI coding tools.
All missed by their existing security pipeline.
All fixed with exact one-line commands.
The tool is called VibeGuard.
What VibeGuard Is
VibeGuard is a free open-source security linter
built specifically for AI-generated code.
Not a generic SAST tool. Not another Bandit wrapper.
Every rule was written by studying actual
AI-generated code and cataloguing the exact
vulnerability patterns these tools produce.
Here is what it catches:
→ SQL injection via f-strings
Copilot writes f"SELECT * FROM users WHERE id = {user_id}"
Looks clean. Works perfectly.
Also lets any attacker dump your entire database.
→ Hardcoded API keys and passwords
AI tools write secrets directly into source code
because that is what training examples looked like.
→ JWT without algorithm verification
ChatGPT generates jwt.decode(token, secret)
Missing the algorithms parameter.
Vulnerable to the alg:none bypass attack.
→ Command injection via shell=True
subprocess.run(cmd, shell=True) everywhere.
User input reaches the shell. Game over.
→ MD5 for password hashing
Still appearing in AI-generated auth code in 2026.
Crackable in seconds.
→ DEBUG=True shipped to production
Every AI tool sets this by default.
Exposes your entire stack to anyone who triggers an error.
What Makes It Different
Every finding includes the exact working code
to fix the problem.
Not a description.
Not a link to documentation.
The actual replacement code.
Copy. Paste. Fixed.
How To Run It
git clone https://github.com/zeroFhacker/vibeguard
cd vibeguard
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt
PYTHONPATH=. python -m vibeguard.cli scan --path ./your-project
Works on Python, JavaScript and TypeScript.
Zero configuration.
Grades your code A to F.
The Result On My Own Project
Grade F. 3 critical findings.
In files I thought were secure.
That is the point.
AI tools are fast. They are powerful.
They will not slow down.
But they need a security layer between
what they generate and what ships to production.
VibeGuard is that layer.
It Is Open Source
MIT licensed. Zero cloud. Zero tracking.
Everything runs locally on your machine.
Contributions welcome — especially rules for
Go, Rust, Java and infrastructure as code.
🔗 github.com/zeroFhacker/vibeguard
Run it on your codebase before the next deployment.
It takes 30 seconds.
It is free.
It might save you from a very bad day.
Built by David Seagal — security engineer and
open source builder at github.com/zeroFhacker
Top comments (0)