Meta 论文提出 TextSeal2：双密钥合一提升 AI 文本水印检测率
New Meta paper shows that blending 2 secret keys into 1 makes hidden watermarks in AI-written text easier to detect, even when the AI uses a common trick (like
Meta 新论文提出 TextSeal2，把两个密钥融合为一个虚拟密钥并加入新噪声，使每个 token 都携带完整水印信号，同时保持答案多样性。针对投机解码，它从草稿随机竞争的剩余项中选取替换 token，使接受率不变。在 Qwen3.5-9B 上启用投机解码且多样性相同时，TextSeal2 在 0.1% 误报率下对 200 token 回答的检测率为 89%，TextSeal 为 71%，textGrain 为 67%。
推荐理由
AI 评分 55
Meta 新论文把双密钥合为单一虚拟密钥，在投机解码下仍保持检测信号，并给出与 TextSeal、textGrain 的同条件对比。
主题
正文
New Meta paper shows that blending 2 secret keys into 1 makes hidden watermarks in AI-written text easier to detect, even when the AI uses a common trick (like speculative decoding) to write faster.
Watermarks hide a secret-key signal in which tokens an LLM samples, so a detector can later flag its text. Meta's earlier TextSeal picked 1 of 2 keys per token at random, so detection had to average both and lost signal.
TextSeal2 blends both keys into 1 virtual key plus fresh noise, so answers stay varied and every token carries full signal. For speculative decoding, it picks replacement tokens from the leftovers of the drafts' random races, so acceptance rates stay unchanged.
On Qwen3.5-9B with speculative decoding and equal diversity, it detected 89% of 200-token answers at a 0.1% false-positive rate. TextSeal caught 71% and textGrain 67%, at the same acceptance rate.