Meta 新论文 TextSeal2：双密钥融合让 AI 文本水印更易检测
Meta 把两个水印密钥合成一个，speculative decoding 下也能检出 89% 的 AI 文本，比旧版 TextSeal 高了 18 个百分点。
Meta 发表新论文 TextSeal2，改进其此前 TextSeal 水印方案。原方案在两个密钥中按 token 随机选一，检测时需对两者取平均而损失信号；新方案将两个密钥融合为一个虚拟密钥并加入噪声，每个 token 都携带完整信号。针对 speculative decoding 场景，它从草稿随机竞争的剩余 token 中挑选替换词，保持接受率不变。在 Qwen3.5-9B 上配合 speculative decoding 测试，TextSeal2 在 0.1% 误报率下对 200-token 回答检出 89%，高于 TextSeal 的 71% 和 textGrain 的 67%。
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.
