ClockMark: Adaptive and Host-Dependent Watermarking for Large Language Models
Large language model (LLM) watermarking operates on the next-token distribution that evolves during generation.
ProofPaper ↗
Key points
- To accommodate this difference, we formulate a novel embedding framework for LLM watermarking based on two principles.
- We further argue that embedding should be host-dependent, using the current LLM distribution to constrain both token selection and state updates.
- Guided by these principles and the two requirements of distribution preservation and robustness to text editing, we propose ClockMark.
- Theoretically, we prove that ClockMark preserves the original LLM sequence distribution, thereby maintaining generation quality.
Sources (1)
- [1]ClockMark: Adaptive and Host-Dependent Watermarking for Large Language ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 06:00 PM
Large language model (LLM) watermarking operates on the next-token distribution that evolves during generation.
To accommodate this difference, we formulate a novel embedding framework for LLM watermarking based on two principles.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 19, 2026Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM
- Oct 19, 2026From Black Box to Controller: Inference-Time Steering of LLM via Feature Amplification
- Oct 18, 2026System X: An LLM Decision Model That Replies from Its Own KV Cache
- Oct 18, 2026CorrectRTL: Boost RTL Generation Quality Through Automated Self-Correction with Instruction Decomposition
- Oct 17, 2026Pair-Aware Optimization: Coordinating Exploration and Refinement for Jailbreaking LLM Agents
- Oct 16, 2026CME-SpectrumBench: Can LLMs Analyze Condensed Matter Spectral Data?