In ours -- using any sensible definition of intelligence and will. Unfortunately, unless they analyze themselves quite closely, people that ask this question tend to automatically adopt definitions that are performance-oriented and not psychologically astute.
Why can't an AI have free will? You seem to have taken this as a principle without any reasoning. And the definitions that tend to lie behind this principle don't match well with human behavior. Primarily: we don't consider it mere slaughter, but actual murder, to kill the stupid.
From a psychodynamic point of view, a computational mechanism would only qualify as an intelligence if it can derive agenda of its own, independent of the agenda supplied to it. Note that this has nothing to do with a level of performance. There is often a confusion between two very different uses of the word 'intelligence', and the performance-oriented one simply does not belong in this context. We do not, after all, consider children or those with low IQ's to be less than human, or to entirely lack minds. They remain sentient, conscious, intelligences -- just of limited capacity.
But we do consider all tools to lack intelligence -- as tools they merely replicate and apply our goals and beliefs, they do not have their own. We even call a human who attempts complete submission to some authority a tool, implying a purposeful disavowal of their real intelligence. So intelligence is not identified primarily by logical capacity, and never has been. Free will is not about logic, it is about will.
More practically, in order to pass that Turning test, the intelligence would have to be able to display a certain level of capriciousness, or it would not seem human. So passing a Turing test may require indeterminacy, and that indeterminacy would need to extend to its goals and objectives. Otherwise, it would not be psychologically convincing over long periods of time. Again, free will is not about logic, it is about will.
I just realized that I have not answered the question. The answer that flows out of these obervations is that free will is possible where agenda matters in any way. That means that when determining forces are balanced, or close enough to being balanced, there are multiple outcomes allowed.
(If when forces are balanced, there is still only one outcome, you have a very strange set of forces that preclude the possibilities of zeroes, and thus don't obey the laws of mathematical combination -- you live in the world of Cauchy's infinitesimals, where zero is never really zero, but is some tendency toward some number, despite having no magnitude. Of course, absolute balance is rare, but if when forces are still unbalanced, but very close to being balanced, there is still only one outcome, you have rejected the observations of quantum mechanics.)
Another way of putting it is that free will is possible in constructions where evolution actually searches a solution space, (i.e. actually serves a purpose.) The fact of genetic and social evolution, and the fact that it seems not to be a uniform process, but one that creates novel directions at unexpected intervals suggests that novel solutions are possible -- that balanced-but-large forces do not result in indefinite immobility or stagnation, but in multiple possible outcomes.