AI Models Write Code with Security Flaws 18–50% of the Time | Not Hacker News!