There's been little improvement in how well AI models handle core security decisions, says a report from application security ...