One of the problems with this description in general is that virtual particles aren't necessarily actual things that ever exist.
Virtual particles stem from a particular method for calculating interactions in what is known as scattering theory. Scattering theory involves some number of particles entering from infinity, interacting with one another in some way, and then some number of particles exiting out to infinity. The "scattering amplitude" of a given outcome tells you how likely that outcome is. For example, if you have an electron and a positron collide, there is some probability for them to annihilate into a pair of photons. There is also some probability that they will, if they have enough energy, produce a pair of muons. Or tau particles. Or neutrinos. Given the energies of the incoming particles, you can calculate the probability of each of these using scattering theory.
The problem is: nobody knows how to solve the equations involved in that calculation exactly. We have to use approximations. And the primary approximation that is used is known as perturbation theory. The idea of perturbation theory is that rather than trying to solve the full equation, we break it up into distinct, solvable pieces which can be solved individually and added together. Mathematically, these distinct pieces end up looking a lot like the creation and annihilation of particles: the calculation, once performed, looks like when the electron and positron come together, there are a whole bunch of photons, electrons, positrons, and other more exotic particles which pop into existence for a fraction of a second before annihilating.
But there's a problem: it doesn't look exactly like the above situation. The masses of the temporary (i.e. virtual) particles are all wrong: while a photon has zero mass, the virtual photon which appears in the above calculation has imaginary mass (that is, the square of its mass is negative). It's not even all that hard to prove that virtual photons must have imaginary masses (this fact draws directly from conservation of momentum and energy of the incoming and outgoing particles). The fact that these virtual particles don't have the right masses may be a hint that this perturbation theory approach isn't actually describing anything that actually happens: it may just be a coincidence that the calculation looks sort of like the creation/annihilation of particles.
So, with the black hole case, we know that the outgoing particles necessarily have positive energy (as PeterDonis described). As the theory respects local energy conservation, that energy has to come from the black hole. Thus, if the system is described as virtual particles falling into the black hole, then those virtual particles must have negative energy in order to satisfy the conservation laws that the equations require. Note that this conservation law isn't anything imposed externally: it falls out of the equations. But given that virtual particles may be mathematical figments anyway, it's not worth taking that kind of description too seriously.