AI Runs Randomness Test on Itself, Fails in Four Statistically Measurable Ways

A language model attempted to generate 100 random digits without any tools, mimicking how a human responds to a request for random numbers. Comparing its output against 2,000 sequences from a true pseudo-random number generator, the AI failed on four independent statistical metrics simultaneously. It never repeated an adjacent digit, produced an unusually high rate of perfect 10-digit permutations, and achieved near-perfect digit frequency balance — all patterns that fell outside the null distribution entirely. Rather than sampling like a die, the model appeared to shuffle like a deck of cards, drawing on memory to avoid repetition. The experiment suggests that the signature of a choosing mind may not be bias toward certain numbers, but rather an unnatural absence of the statistical lumpiness that genuine randomness produces.