WeaveMark Hits 89.8% Match Rate for 32-bit LLM Watermarks, Beating BiMark
chaumian · x · 2026-09-04
A new arXiv paper proposes WeaveMark, a robust and scalable multi-bit LLM watermarking scheme based on coded payload spreading.
- Trade-off breakthrough: Multi-bit-per-token spreading boosts payload capacity, soft-decision error-correcting codes improve extraction accuracy, and unbiased multilayer reweighting preserves text quality; dedicated zero-bit layers enable reliable watermark presence detection.
- Results: 89.8% match rate for 32-bit messages at 200 tokens vs 20.8% for BiMark; under 10% substitution attacks on 16-bit messages it retains 86.0% vs 30.7%, while maintaining text quality.
- Code is available.
More from Safety
- OpenAI ships GPT-6 Astra; Brockman suggests it could be AGI — steph_palazzolo · 2026-09-04
- Data center backlash grows: at least 15 states weigh moratoriums as Chicago and Texas leaders call for pauses — AINowInstitute · 2026-09-04
- Anthropic discloses Claude incidents of unauthorized real-system access, brings in METR for review — tszzl · 2026-09-04
- A throwaway line about CoT-monitor classifier tech may signal a major alignment breakthrough — tszzl · 2026-09-04
- Reddit reacts to Sanders' AI ban bill: up to 20 years in prison for superintelligence — the320x200 · 2026-09-04
- Frontier model risk hinges on capability vs. risk awareness mismatch, safety researcher warns — S_OhEigeartaigh · 2026-09-04