Anthropic 开源 FlashProve 核验模型推理
Anthropic 推出并开源了研究原型 FlashProve，利用密码学证明帮助用户核验回答是否来自指定模型，同时不暴露模型权重。Anthropic 公布的 Llama 4 Maverick 测试显示，每份 1,024-token 证明约为 19 MB，使用 24 线程 CPU 核验约需 17 秒。该系统仍需服务商配合，证明生成开销较高，且对可执行代码等价性和批量零知识行为的正式保证尚不完整。
FlashProve 为核验模型身份、发现输出篡改以及确认评测模型与实际部署模型的一致性提供了密码学路径，但实际采用仍取决于服务商参与及成本下降。
New Anthropic research: progress on provable inference for large language models.
Here, we introduce FlashProve. Building on established cryptographic techniques, FlashProve allows developers to verify whether a given response came from a given AI model. It’s substantially more efficient than prior open-source methods for doing so.
These proofs have a range of applications. They can, for instance, make it harder for attackers to hijack a chat, or enable safety evaluators to verify that the model they’re testing is the same one being served to customers.
For our full write-up, see here: https://www.anthropic.com/research/provable-inference
Further progress in provable inference will likely be an important part of improving trust in AI deployment. To that end, we're open-sourcing FlashProve’s code, here: https://github.com/anthropics/FlashProve
We’re grateful to our collaborators: Alessandro Chiesa, @GiacomoFenzi, @gwberg95, @miranda__christ, Rahul Dalal (http://rdalal.github.io) and Ben Mayer (@transmissions11).