It could still be a pseudo random number generator behind the scenes. For example, a typical quantum circuit simulator would implement measurements by computing a probability then asking a pseudo random number generator for the outcome and then updating the state to be consistent with this outcome. Bell's theorem proves those state updates can't be local in a certain technical sense, but the program has arbitrary control over all amplitudes of the wavefunction so that's not a problem when writing the simulator code.

If the prng was weak, then the quantum circuit being simulated could be a series of operations that solve for the seed being used by the simulator. At which point collapses would be predictable. Also, it would become possible to do limited FTL communication. An analogy is some people built a redstone computer in minecraft that would detonate TNT repeatedly, record the random directions objects were thrown, and solve for the prng's seed [1]. By solving at two times, you can determine how many calls to the prng had occurred, and so get a global count of various actions (like breaking a block) regardless of where they happened in the world.

[1]: https://www.youtube.com/watch?v=FPmQ0rnJjNc