Assuming the nature of the simulation is entirely in the hands of its designers to specify, the rules of that simulation could readily include a one-way mechanism for information injection that, again by design, we'd never be able to identify.
Only the simulated entities within the simulation (i.e. us and everything else we perceive) are fully subject to the entire rules of the simulation, the same is not true for something outside trying to debug it :-D