AI & ML
Watermark Evidence under Open-Weight LLM Modification: Exact Bounds for Low-Rank Output Watermarks
arXiv cs.CR0 views
arXiv:2611.21702v1 Announce Type: new Abstract: Open-weight language models can continue generating after their released weights are modified, while a watermark verifier still scores outputs against the original base and marked models. We study how such modifications change the evidence available to an OpenStamp-style verifier. At each shared prefix, we represent the modification as a residual in the next-token logits. We derive an exact finite-vocabulary lower bound on the expected log-likelihood-ratio evidence under a residual-span constraint, along with a sharp span-only deviation bound. We also characterize the first-order evidence shift through a covariance term with an explicit second-order remainder, and determine the residual span needed to reach a target evidence level. For sequence-level detection, we derive an exact conditional exponential-moment bound and combine it with a supermartingale argument and a distribution-free order-statistic tolerance bound. This yields a high-confidence upper bound on miss probability after attack selection on calibration data and evaluation on independent certification continuations. In experiments with OpenStamp on Llama-2-7B and Mistral-7B-v0.3, INT8 quantization and 30% pruning retain test true-positive rates of 85.6-94.8%, while 500-step LoRA fine-tuning lowers them to 28.8% and 11.6%. The covariance prediction tracks the measured mean LoRA evidence drift on both models, with no remainder-bound violations across 160,000 test prefixes. Additionally, we observe zero false positives among 500 base-model continuations per model. Through these theoretical and empirical results, the current research provides a better understanding of watermark robustness to model-editing/tuning attacks.
Source:https://arxiv.org/abs/2611.21702PreviousStop Early, Stay Simple: Evidence Acquisition for Power System Fault and Attack DiagnosisNextLLMs as Validators for Cross-Chain Bridges: Evaluating Source-Level and Transaction-Level Attack Detection on Solana
Related posts
What does one finished AI task cost? I measured it.
0dev.to AI
Suno music platform gets new feature for image and video creation
0dev.to AI
How to Trim Audio for a Comedy Show Video — Free, in Your Browser
0dev.to WebDev
AI Internship for Students in Bangalore: Develop Practical Skills and Explore Career Opportunities
0dev.to WebDev
SPF PermError: too many DNS lookups
0dev.to Security

Shipping a 1999 strategy game on Linux ARM64: static linking, SHA-256 installers and 40 new pixel-art portraits (JA2 v1.13)
0dev.to Linux