The deeper problem here isn't that deepfakes are too good - it's that every "proof of humanity" test converges to the same bag of tricks. Shared secrets, liveness checks, biometric challenges. An attacker who studies the test can pass it. We keep building Voight-Kampff machines without asking whether the Voight-Kampff framing is the right one. The question isn't "can you tell this is real" - it's "what would you accept as proof, and can that proof be synthesized?"

LLM slop. See their comment history.