Event ·Official first-hand

Anthropic 发布 FlashProve：用密码学证明回复来自指定模型

Anthropic 开源了 FlashProve，一个基于 zkSNARK 的原型系统，能验证回复确实来自被承诺的模型权重，同时不泄露权重本身。项目采用 Apache 2.0 许可证。在测试中，它对 Llama 4 Maverick 的推理证明达到每秒 110 个 token，运行在单块 H100 上。单个证明大小为 19 MB。

Current read

Anthropic 开源了个挺有意思的东西：不泄露模型权重，也能用密码学证明回复真是这个模型生成的，Llama 4 上能跑到每秒 110 token。

11 sources51° AI heatLast updated 2026/11/3 22:08:22

Evidence

11 sources

Primary sourcemarktechpost

2026/11/3 22:08:22View source

Related source 1AIGCLINK

2026/11/4 01:08:31View source

Related source 2Gorden Sun

2026/11/3 08:00:09View source

Related source 3rohanpaul_ai

2026/11/3 20:15:49View source

Related source 4Anthropic

2026/11/3 20:10:08View source

Related source 5techcrunch

2026/11/3 22:41:03View source

Related source 6shao__meng

2026/11/2 04:06:38View source

Related source 7宝玉

2026/11/2 19:34:14View source

Related source 8Pieter Levels

2026/11/2 21:55:09View source

Related source 9Blognone

2026/11/3 02:25:26View source

Related source 10IT之家

2026/11/3 07:43:03View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story