Anthropic研发并开源AI模型验证系统FlashProve

1. 16:04
   IT之家 · AI 筛选

   Anthropic 正在开发开源系统 FlashProve，通过密码学验证 AI 输出来源模型

   Anthropic 正在开发名为 FlashProve 的开源系统，旨在通过密码学证明为 AI 输出提供数学层面的保证，以验证内容确实由指定模型生成。该系统利用不泄露内部权重的模型指纹与输出进行匹配，支持在本地轻量验证，可防止厂商偷换低成本模型或伪造测试结果。目前相关技术规范与代码已在 GitHub 开源，但仍处于阶段性研究阶段，尚未正式部署。

2. 11:28
   AITNT · AI头条

   Anthropic 开源 FlashProve 可验证推理系统

   Anthropic 开源大模型可验证推理系统 FlashProve，基于零知识证明技术，允许服务商在不公开模型权重的前提下证明回答确实由指定模型生成。在 400B 参数的 MoE 模型测试中，单张 NVIDIA H100 GPU 证明生成吞吐量达到约 110 Token/s，证明文件约 18.78 MB，24 线程 CPU 验证约需 17 秒。

3. 06:41
   TechCrunch AISelected

   Anthropic develops open-source system FlashProve to verify AI model authenticity

   Anthropic is developing an open-source system called FlashProve that uses verifiable inference to provide cryptographic proofs for model outputs, ensuring users can verify that responses originate from a specific model. The system requires AI labs to publish model fingerprints that do not expose weights, enabling lightweight local verification to prevent vendors from covertly swapping in cheaper models to cut costs or inflating benchmark results. Its technical specifications and code have been open-sourced on GitHub, though the project remains work-in-progress research and has not yet been deployed in production.

Current heat 25·Comparable peak 26（Nov 4 17:00）·Comparable change, last 24 hours–

Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.