Keep your generation process,
shred the watermark.
Shredders intelligently diversify long-form text through dozens of models to remove high-confidence watermarks. Proven to work.
Multi-model kernels · Reads like a human · Compliant by construction
The watermark falls below detection
We score Lettershred output against a KGW green-list detector, the same watermark family the labs ship. Watermarks register when z hits 4. Shredders keep it under.
z fell from 15.46 to −0.32, below the detection threshold of 4. The text now scores like it was never watermarked.
177 of 240 tokens rewrittenThe hairline marks the detection threshold (z = 4). Only the watermarked run clears it.
The sweep — mean z over 24 seeds at 150 tokens. Read the last two columns as the control: they rewrite just as much text, but blind to the key or in blocks. Evasion comes from diversifying across models and dispersing the rewrites, not from changing more words.
| Coverage | Cross-provider, dispersed | Cross-provider, blocked | Re-imprint, dispersed |
|---|---|---|---|
| 0% | 12.21 | 12.21 | 12.21 |
| 25% | 6.36 | 9.06 | 9.21 |
| 50% | 2.14 | 5.41 | 6.18 |
| 75% | -0.32 | 2.87 | 4.09 |
Beat the watermark
without changing your process.
One brief in, one polished, multi-model article out.