English

ForgetMark: Stealthy Fingerprint Embedding via Targeted Unlearning in Language Models

Cryptography and Security 2026-01-21 v2 Artificial Intelligence

Abstract

Existing invasive (backdoor) fingerprints suffer from high-perplexity triggers that are easily filtered, fixed response patterns exposed by heuristic detectors, and spurious activations on benign inputs. We introduce \textsc{ForgetMark}, a stealthy fingerprinting framework that encodes provenance via targeted unlearning. It builds a compact, human-readable key--value set with an assistant model and predictive-entropy ranking, then trains lightweight LoRA adapters to suppress the original values on their keys while preserving general capabilities. Ownership is verified under black/gray-box access by aggregating likelihood and semantic evidence into a fingerprint success rate. By relying on probabilistic forgetting traces rather than fixed trigger--response patterns, \textsc{ForgetMark} avoids high-perplexity triggers, reduces detectability, and lowers false triggers. Across diverse architectures and settings, it achieves 100\% ownership verification on fingerprinted models while maintaining standard performance, surpasses backdoor baselines in stealthiness and robustness to model merging, and remains effective under moderate incremental fine-tuning. Our code and data are available at \href{https://github.com/Xuzhenhua55/ForgetMark}{https://github.com/Xuzhenhua55/ForgetMark}.

Keywords

Cite

@article{arxiv.2601.08189,
  title  = {ForgetMark: Stealthy Fingerprint Embedding via Targeted Unlearning in Language Models},
  author = {Zhenhua Xu and Haobo Zhang and Zhebo Wang and Qichen Liu and Haitao Xu and Wenpeng Xing and Meng Han},
  journal= {arXiv preprint arXiv:2601.08189},
  year   = {2026}
}

Comments

Accepted by ICASSP2026

R2 v1 2026-07-01T09:02:04.513Z