# Claude 文本水印机制：密钥掷硬币检测 AI 生成

- 来源：Dongxi 东锡 NLP (@dongxi_nlp)
- 发布时间：2026-08-15 03:45
- AIHOT 分数：44
- AIHOT 链接：https://aihot.virxact.com/items/cmstdjoio044hro0xbzunnerx
- 原文链接：https://x.com/dongxi_nlp/status/2088351263453065675

## AI 摘要

马东锡 NLP 解析 Claude 的 SynthID-Text 水印机制：模型前向传播后，对每个候选 token 进行一次带密钥的上下文相关“掷硬币”更新，再采样生成。持有密钥者可检测该统计模式。Anthropic 称此举为遵守欧盟 AI 法案，不影响输出质量、不增加成本，且水印无法追溯到具体个人或对话。

## 正文

The SynthID-Text watermarking LLM generation is like:

model forward pass
-> next-token logits
->temperature / top-k / top-p
---> keyed watermark probability update
->sample a token

For every plausible next token, Claude consults a secret, context-specific coin flip.

"Someone" holding the secret key can measure that pattern.

### 引用推文

> Anthropic：We've written an FAQ to answer some of the questions we've received about watermarking. In summary: • We're implementing watermarking to comply with the EU AI A...