Research map / Large language models / Language, safety and interpretability
De-Anonymizing Text by Fingerprinting Language Generation
Zhen Sun, Roei Schuster, Vitaly Shmatikov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use sequences of length N>=1900 instead of shorter | fingerprint attack · reddit-sports | recall: greater than 99\% | Improved |
| remove 6% noisiest traces by noise level (ablation) | fingerprint attack · reddit-sports | measurement error outliers: 6\% | Improved |
| accounting for the 6% filtered traces | attack recall · reddit-sports | recall: 99\% -> 93\% | Hurt |
| apply attack to short chat posts (avg 9.5 words) instead of long forum posts | fingerprint attack · Ubuntu Chat | recall: 93.4\% | Mixed |
| restrict Silk Road users to those with variable nucleus sequences | fingerprint attack · Silk Road forum | users de-anonymized without false positives or negatives: 18 out of 41 | Improved |
| replace data-dependent top-p filtering with fixed-iteration constant-flow algorithm | auto-completion sampler · loop execution time | slowdown: 1.15x | Hurt |
| fixed-iteration mitigation | sample_sequence · 2700-word auto-completion simulation | runtime: 0.3\% | No clear effect |
| multiply measured iteration count by 100/1.1 to correct missed accesses | Flush+Reload attacker · victim loop iterations | iteration estimate: 1.1\% | Improved |
Closest papers by meaning
- Self-Refining Language Model Anonymizers via Adversarial Distillation (NeurIPS 2025 · similarity 0.85)
- Language Models are Advanced Anonymizers (ICLR 2025 · similarity 0.84)
- Humanizing the Machine: Proxy Attacks to Mislead LLM Detectors (ICLR 2025 · similarity 0.84)
- Fingerprinting Inference Systems of Large Language Models (NeurIPS 2026 · similarity 0.84)
- Scalable Fingerprinting of Large Language Models (NeurIPS 2025 · similarity 0.84)
- InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
