Anyone holding out hope that AI can be made inherently secure are in for a disappointment, with Microsoft's research team saying it is an impossible task.