AI Models Write Code with Security Flaws 18–50% of the Time

A study found that AI models write code with security flaws 18-50% of the time, prompting discussion on how this compares to human-written code.

Snapshot generated from the HN discussion