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

What is the rate of security flaws in code written by humans? Difficult to asses the results without a base rate for comparison.