Are You Human? An Adversarial Benchmark to Expose LLMs