← Wednesday, November 4, 2026
AI Frontier · Nov 3
LowAnthropic builds a way to prove which AI really answered you

Anthropic makes the AI tool Claude. It shared a new system called FlashProve. It can prove that an AI answer came from the exact model it claims to be. Each model gets a kind of fingerprint. Later, anyone can check an answer against it. This does not give away the model's secrets. It is early work, not a finished product.
Why it matters to GCs
You may soon trust AI for takeoffs, RFIs, or code checks. This is about knowing the answer really came from the strong model you paid for, not a weaker swap. For risky work, proof of what did the job helps with trust, and with blame if something goes wrong.
TechCrunch·Via POLR AI Hub