Invisible Ink, Visible Lies: How Production Watermarking Causes LLMs to Hallucinate

cs.CR, cs.AI

Submitted: 2026-10-04

Updated: 2026-10-04

Comments: Accepted at NeurIPS 2026

Code: https://github.com/Haocheng-NJUST/WM-Hallucination

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Sources

• Retrieval-Augmented Generation for Large Language Models: A Survey
• Towards Trustworthy Retrieval Augmented Generation for Large Language Models: A Survey
• A Resilient and Accessible Distribution-Preserving Watermark for Large Language Models
• The Llama 3 Herd of Models
• Mistral 7B
• Qwen2.5 Technical Report
• TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection
• MorphMark: Flexible Adaptive Watermarking for Large Language Models

Related papers

• SoK: AI-Augmented Binary Reversing
• Relaxed Sender Anonymity for CBDC Interbank Settlement: A Zero-Knowledge Approach on Permissioned EVM
• Calibration-Family Overfit: Why Trusted Sabotage Monitors Don't Transfer Across Lineages
• Efficient Fuzzy PSI under One-Sided Assumptions
• Sealing the Audit-Runtime Gap for LLM Skills
• Token Composition: A Graph Based on EVM Logs