The use of AI in legal decision making – an incipient yet seemingly inevitable phenomenon – presents some problems. Some of them are of pure technical nature (e.g. the ‘coding of progressively more sophisticated AI algorithms, their adaptability, their reliability, etc.). Others are more theoretical, involving social and political issues (questions of transparency, accountability, bias, ownership, etc.). Both sets of problems are challenging in their own ways, but they seem, at least in principle, susceptible to solutions compatible with the fundamental tenets of modern law. In this sense, they can be labelled as ‘Easy Problems’: even if they will be eventually unsolvable, at least their cognitive logic is clear to us. There is, however, another problem posed by AI that I propose to label as the ‘Hard Problem’ of AI, because it presents a more fundamental challenge for modern law. Simply stated, my claim is that if human dignity – which is as the core of our legal universe – is to be understood, as several authors do, as the capacity by individuals to advance claims about themselves, then AI is fundamentally undermining the legitimacy of those claims. If AI can read, analyze, and predict reality better than any human can do, then AI knows better about humans too. It follows that when people’s claims clash with AI determinations, it is AI, with its ‘omniscience’, that must have the upper hand. The realisation of the ‘Hard Problem’ brings us to question the limits of the applicability of AI to the legal field or, perhaps, to question the fundamental structures of modern law itself.