Interesting experiment even though it's still early days for this kind of research and we'd need more such evaluations. They found that it produces mostly insecure code. This is not really surprising in the end, this manipulates language but has not execution model. It can be fixed only by coupling to some outside system.