Skip to content
Trending storyDeveloping

Meta's TextSeal2 watermark detects 89% of AI text at a 0.1% false-positive rate

1 article1 sourcesince Nov 2Last article 3h ago ·

Overview

AISummary of 1 article

Meta publishes TextSeal2, a watermarking method for LLM text that blends two secret keys into one virtual key plus noise, so every token carries full watermark signal instead of half.

According to Rohan Paul's summary of the paper, under speculative decoding TextSeal2 draws replacement tokens from leftover draft samples, keeping acceptance rates unchanged.

On Qwen3.5-9B, the paper reports detecting 89% of 200-token answers at a 0.1% false-positive rate, compared with 71% for TextSeal and 67% for textGrain.

Written by AI from the articles below · updated Nov 2, 10:39 AM ET

Check the sources:

Latest developmentNov 2, 10:20 AM ET
Meta's TextSeal2 improves AI text watermark detection under speculative decoding

Article timeline

The articles in this story. Times are ET.

Nov 2
  1. Rohan PaulX
    Meta's TextSeal2 improves AI text watermark detection under speculative decoding

    AIMeta's TextSeal2 blends two secret keys into one virtual key plus noise, so every token carries full watermark signal instead of half. With speculative decoding, it draws replacement tokens from leftover draft samples, keeping acceptance rates unchanged. On Qwen3.5-9B, it detects 89% of 200-token answers at a 0.1% false-positive rate, versus 71% for TextSeal and 67% for textGrain.

    Image from @rohanpaul_ai's post

Heat Trend

Current comparable heat
9
Peak comparable heat
10Nov 2, 11:00 AM ET
Change in past 24 hours
–

The trend compares only continuously observed participants, so its scope may be smaller than the current heat total. Hover or tap for hourly heat; use arrow keys with a keyboard.