Watermark Edit Detection Error-correcting-code watermarks for detecting insertions, deletions, and substitutions after generation. Anchor-ECC Localizing suspicious regions in edited watermarked LLM outputs. Watermarkability Measuring and improving model compatibility with logit-based watermarks. HawkesLLM Semantic uncertainty propagation in agentic text simulation. Robust Inference for Delayed Point Processes Correction-centered distributional robustness under reporting delays and model misspecification.
