This article examines reconstructability as a threshold condition in AI-mediated enterprise contexts. It does not argue that courts, regulators, or enterprises require AI outputs to be accurate, explainable, or deterministically reproducible in all cases. Instead, it isolates a narrower and prior question: whether an enterprise can reconstruct what representation was generated, when it was generated, and under what system conditions once scrutiny arises.