Newest papers

Every Model Leaves a Trace: Multi-Resolution LLM Fingerprinting

Mohammad Reza Modarres, Armin Tourajmehr, Artin Tavasoli, Yadollah Yaghoobzadeh, Mohammad Taher Pilehvar · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28434 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Model bahasa besar meninggalkan jejak spesifik model dalam teks mereka yang dihasilkan, tetapi tetap tidak jelas seberapa halus sidik jari ini dapat membedakan model terkait dan sinyal mana yang tetap informatif karena petunjuk yang jelas dikurangi. Kami mempelajari ini melalui resolusi sidik jari, mempertimbangkan kedua resolusi model, dari keluarga yang berbeda untuk varian distilasi dan disesuaikan, dan resolusi sinyal di seluruh perilaku, stylometer, leksikal, dan bukti semantik. Kami secara progresif menekan petunjuk permukaan, output parafrase, bervariasi jumlah teks yang tersedia, dan tes generalisasi di seluruh distribusi instan dan bahasa. Hasil kami menunjukkan bahwa sidik jari bertahan di luar perbedaan keluarga yang luas, tetap terdeteksi dalam model distilasi, dan bahkan adaptasi LoRA dari basis model yang sama. Meskipun atribusi melemahkan permukaan

Machine translation (Meta M2M100); the English original follows.

Abstract

Large language models leave model-specific traces in their generated text, but it remains unclear how finely these fingerprints can distinguish related models and which signals remain informative as obvious cues are reduced. We study this through fingerprint resolution, considering both model resolution, from different families to distilled and adapted variants, and signal resolution across behavioral, stylometric, lexical, and semantic evidence. We progressively suppress surface cues, paraphrase outputs, vary the amount of available text, and test generalization across prompt distributions and languages. Our results show that fingerprints persist beyond broad family differences, remaining detectable in distilled models and even LoRA adaptations of the same base model. Although attribution weakens as surface and lexical cues are altered, substantial model-specific signal remains. These signals transfer across prompt sources and languages, with semantic representations proving most robust across languages. These findings show that model identity is recoverable at multiple resolutions, but that recoverability depends on model similarity and the evidence available in the text.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Paraphrase, Artificial intelligence, Generalization, Natural language processing, Computer science, Similarity (geometry)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange