5:32 PM ET · Nov 4

Anthropic open-sources FlashProve, a cryptographic system to prove which AI model wrote a response

Anthropic released FlashProve, an open-source research prototype that uses zkSNARK cryptographic proofs to show a reply came from a specific committed model without revealing the model's weights. On Meta's 400B-parameter Llama 4 Maverick, a single Nvidia H100 generates proofs at about 110 tokens per second, and each ~19 MB proof can be checked on an ordinary CPU in under half a minute. Anthropic says the approach could stop providers or routers from quietly swapping in cheaper models and let regulators verify which model is being served, but the prototype handles only 1,024-token contexts and greedy decoding, and there are no plans to deploy it yet.

Source: TechCrunch

On X

Huge congrats to Succinct’s Rahul Dalal, one of the cryptographers behind FlashProve, @AnthropicAI’s provable inference system built on zkSNARKs. Why: Today, AI runs on a “just trust me” model. Users can’t verify that a response came from the model they chose, or that an
— Succinct (@SuccinctLabs) Nov 3, 2026

Last attempt at proving which model wrote a response: 14 hours and a 93 GB proof. Anthropic’s new one: a few seconds and 19 MB. It’s called FlashProve. Right now you can’t be sure which model answered your prompt. A provider could quietly swap in a cheaper one, or a hacker
— Pratham (@Prathkum) Nov 3, 2026

.@AnthropicAI just open-sourced FlashProve, a zkSNARK system for verifying that an AI response was generated by a specific set of model weights, without revealing those weights. The idea is simple: today, when you access an LLM through an API or a third-party router you're
— 100y (@100y_eth) Nov 3, 2026

Back to The Wire · Subscribe

Anthropic open-sources FlashProve, a cryptographic system to prove which AI model wrote a response

On X

Know what matters in AI.