Unsurprising if the training set includes loads of code with security vulnerabilities. A more interesting question would be how likely code AI is to introduce

Unsurprising if the training set includes loads of code with security vulnerabilities. A more interesting question would be how likely code AI is to introduce security vulnerabilities with a training set of high quality code that has no (well, hardly) any vulnerabilities. Does it still cause security issues because it doesn't truly "understand" what it does, merely piecing together text fragments with little "thought"?