I Trained a Vulnerability Detection Model. It Was Reading Comments, Not Code.

I trained a simple ML model to detect code vulnerabilities and got 84.7% accuracy. Then I looked at...