Developing

Anthropic open-sources FlashProve, a prototype that proves which model answered

Anthropic published FlashProve on Nov. 2, an Apache 2.0 research prototype that uses zero-knowledge proofs to show a response came from a specific committed model. Tested on Llama 4 Maverick: about 110 tokens/second on one H100, 18.78 MB proofs per 1,024-token request, 70 to 95x GPU overhead. Limited to 1,024-token contexts and greedy decoding.

HEAT
OUTLETS: 3
NEW IN 6H: +2
FIRST SEEN: Nov 3, 2026, 17:40 ET
LAST SIGNAL: 17m ago

Primary source

Anthropicanthropic.com/research/provable-inference

Coverage · 4 articles, oldest first

Tuesday, November 3, 2026

17:40
AnthropicPrimary
Anthropic releases FlashProve, an open-source prototype that cryptographically verifies an AI response came from a given model, not one that a router swapped in
17:41
TechCrunch
Anthropic is working on a system to prove AI models’ identity
21:40
RuntimeWire
runtimewire.com/article/anthropic-flashprove-provable-inference-model-identity

Wednesday, November 4, 2026

00:10
Help Net Security
helpnetsecurity.com/2026/11/04/anthropic-flashprove-provable-inference/

ShareX LinkedIn Hacker News Email

← The whole wire