Continual Pretraining on Encrypted Synthetic Data for Privacy-Preserving LLMs
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911368072921088 |
|---|---|
| author | Liu, Honghao Jiang, Xuhui Xu, Chengjin Yang, Cehao Cheng, Yiran Ni, Lionel Guo, Jian |
| author_facet | Liu, Honghao Jiang, Xuhui Xu, Chengjin Yang, Cehao Cheng, Yiran Ni, Lionel Guo, Jian |
| contents | Preserving privacy in sensitive data while pretraining large language models on small, domain-specific corpora presents a significant challenge. In this work, we take an exploratory step toward privacy-preserving continual pretraining by proposing an entity-based framework that synthesizes encrypted training data to protect personally identifiable information (PII). Our approach constructs a weighted entity graph to guide data synthesis and applies deterministic encryption to PII entities, enabling LLMs to encode new knowledge through continual pretraining while granting authorized access to sensitive data through decryption keys. Our results on limited-scale datasets demonstrate that our pretrained models outperform base models and ensure PII security, while exhibiting a modest performance gap compared to models trained on unencrypted synthetic data. We further show that increasing the number of entities and leveraging graph-based synthesis improves model performance, and that encrypted models retain instruction-following capabilities with long retrieved contexts. We discuss the security implications and limitations of deterministic encryption, positioning this work as an initial investigation into the design space of encrypted data pretraining for privacy-preserving LLMs. Our code is available at https://github.com/DataArcTech/SoE. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_05635 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Continual Pretraining on Encrypted Synthetic Data for Privacy-Preserving LLMs Liu, Honghao Jiang, Xuhui Xu, Chengjin Yang, Cehao Cheng, Yiran Ni, Lionel Guo, Jian Cryptography and Security Computation and Language Preserving privacy in sensitive data while pretraining large language models on small, domain-specific corpora presents a significant challenge. In this work, we take an exploratory step toward privacy-preserving continual pretraining by proposing an entity-based framework that synthesizes encrypted training data to protect personally identifiable information (PII). Our approach constructs a weighted entity graph to guide data synthesis and applies deterministic encryption to PII entities, enabling LLMs to encode new knowledge through continual pretraining while granting authorized access to sensitive data through decryption keys. Our results on limited-scale datasets demonstrate that our pretrained models outperform base models and ensure PII security, while exhibiting a modest performance gap compared to models trained on unencrypted synthetic data. We further show that increasing the number of entities and leveraging graph-based synthesis improves model performance, and that encrypted models retain instruction-following capabilities with long retrieved contexts. We discuss the security implications and limitations of deterministic encryption, positioning this work as an initial investigation into the design space of encrypted data pretraining for privacy-preserving LLMs. Our code is available at https://github.com/DataArcTech/SoE. |
| title | Continual Pretraining on Encrypted Synthetic Data for Privacy-Preserving LLMs |
| topic | Cryptography and Security Computation and Language |
| url | https://arxiv.org/abs/2601.05635 |