Anthropic 发布 FlashProve：用密码学证明回复来自指定模型
Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights
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。
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。
Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights
Anthropic released FlashProve, an Apache 2.0 zkSNARK prototype that proves a reply came from a committed model without revealing weights. It proves Llama 4 Maverick at 110 tokens/second on 1 H100, with 19 MB proofs. The post Anthropic Releases FlashProve: Cryptographic Proofs That a Reply Came From Committed Weights appeared first on MarkTechPost .
