EmbTracker: Traceable Black-box Watermarking for Federated Language Models
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866917335590240256 |
|---|---|
| author | Zhao, Haodong Hu, Jinming Bai, Yijie Dong, Tian Du, Wei Zhang, Zhuosheng Chen, Yanjiao Zhu, Haojin Liu, Gongshen |
| author_facet | Zhao, Haodong Hu, Jinming Bai, Yijie Dong, Tian Du, Wei Zhang, Zhuosheng Chen, Yanjiao Zhu, Haojin Liu, Gongshen |
| contents | Federated Language Model (FedLM) allows a collaborative learning without sharing raw data, yet it introduces a critical vulnerability, as every untrustworthy client may leak the received functional model instance. Current watermarking schemes for FedLM often require white-box access and client-side cooperation, providing only group-level proof of ownership rather than individual traceability. We propose EmbTracker, a server-side, traceable black-box watermarking framework specifically designed for FedLMs. EmbTracker achieves black-box verifiability by embedding a backdoor-based watermark detectable through simple API queries. Client-level traceability is realized by injecting unique identity-specific watermarks into the model distributed to each client. In this way, a leaked model can be attributed to a specific culprit, ensuring robustness even against non-cooperative participants. Extensive experiments on various language and vision-language models demonstrate that EmbTracker achieves robust traceability with verification rates near 100\%, high resilience against removal attacks (fine-tuning, pruning, quantization), and negligible impact on primary task performance (typically within 1-2\%). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_12089 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | EmbTracker: Traceable Black-box Watermarking for Federated Language Models Zhao, Haodong Hu, Jinming Bai, Yijie Dong, Tian Du, Wei Zhang, Zhuosheng Chen, Yanjiao Zhu, Haojin Liu, Gongshen Cryptography and Security Federated Language Model (FedLM) allows a collaborative learning without sharing raw data, yet it introduces a critical vulnerability, as every untrustworthy client may leak the received functional model instance. Current watermarking schemes for FedLM often require white-box access and client-side cooperation, providing only group-level proof of ownership rather than individual traceability. We propose EmbTracker, a server-side, traceable black-box watermarking framework specifically designed for FedLMs. EmbTracker achieves black-box verifiability by embedding a backdoor-based watermark detectable through simple API queries. Client-level traceability is realized by injecting unique identity-specific watermarks into the model distributed to each client. In this way, a leaked model can be attributed to a specific culprit, ensuring robustness even against non-cooperative participants. Extensive experiments on various language and vision-language models demonstrate that EmbTracker achieves robust traceability with verification rates near 100\%, high resilience against removal attacks (fine-tuning, pruning, quantization), and negligible impact on primary task performance (typically within 1-2\%). |
| title | EmbTracker: Traceable Black-box Watermarking for Federated Language Models |
| topic | Cryptography and Security |
| url | https://arxiv.org/abs/2603.12089 |