Provably distortion-free watermarking for any generative model via black-box access
serrjoa · x · 2026-11-19
- A new arXiv paper introduces a semantic watermarking algorithm needing only black-box sampling access to a generative model and black-box access to a semantic embedding.
- The watermark is provably distortion-free: averaged over the secret key, it leaves the generator's output distribution unchanged.
- Detection is robust to content-preserving edits (compression, cropping, resizing), and the construction is fully generic — no white-box access to autoregressive text or diffusion internals required. Instantiated for image generation.
More from Research
- LEV: Looped decision model lifts 2.6B accuracy 63.2% to 76.2% with adaptive compute — jiqizhixin · 2026-11-19
- Aurora2-400M-A20M: a sparse MoE pretrained from scratch in 36 hours on one RTX Pro 6000 — Tall_Abrocoma_3533 · 2026-11-19
- Researcher calls METR/Anthropic '2x speedup' RSI analysis statistically dubious — jessi_cata · 2026-11-19
- Linear Probes Could Reveal Which Persona a Base Model Is Playing, Researchers Suggest — cephaloform · 2026-11-19
- ETH's Anymal robot completes 10km autonomous alpine hike with VLM-guided ROAM system — ChongZzZhang · 2026-11-19
- OpenVax codebase grows from 78K to 565K lines in 15 months — iskander · 2026-11-19