Analyzing Leakage of Personally Identifiable Information in Language Models
Open accessThe DOI matches its Crossref record; 33 of the 33 references it cites were found in a registry or an index.
Abstract
Language Models (LMs) have been shown to leak information about training data through sentence-level membership inference and reconstruction attacks. Understanding the risk of LMs leaking Personally Identifiable Information (PII) has received less attention, which can be attributed to the false assumption that dataset curation techniques such as scrubbing are sufficient to prevent PII leakage. Scrubbing techniques reduce but do not prevent the risk of PII leakage: in practice scrubbing is imperfect and must balance the trade-off between minimizing disclosure and preserving the utility of the dataset. On the other hand, it is unclear to which extent algorithmic defenses such as differential privacy, designed to guarantee sentence-or user-level privacy, prevent PII disclosure. In this work, we introduce rigorous game-based definitions for three types of PII leakage via black-box extraction, inference, and reconstruction attacks with only API access to an LM. We empirically evaluate the attacks against GPT-2 models fine-tuned with and without defenses in three domains: case law, health care, and e-mails. Our main contributions are (i) novel attacks that can extract up to 10× more PII sequences than existing attacks, (ii) showing that sentence-level differential privacy reduces the risk of PII disclosure but still leaks about 3% of PII sequences, and (iii) a subtle connection between record-level membership inference and PII reconstruction. Code to reproduce all experiments in the paper is available at https://github.com/microsoft/analysing_pii_leakage.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Language Models are Few-Shot Learners
Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020) · arXiv · doi:10.48550/arxiv.2005.14165 · Cited by 23,372
RoBERTa: A Robustly Optimized BERT Pretraining Approach
Liu, Y., Ott, M., Goyal, N., Du, J., Joshi, M., Chen, D., Levy, O., Lewis, M., Zettlemoyer, L., & Stoyanov, V. (2019) · arXiv · doi:10.48550/arxiv.1907.11692 · Cited by 16,515
Decoupled Weight Decay Regularization
Loshchilov, I., & Hutter, F. (2017) · arXiv · doi:10.48550/arxiv.1711.05101 · Cited by 12,522
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer
Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., & Liu, P. J. (2019) · arXiv · doi:10.48550/arxiv.1910.10683 · Cited by 10,929
Calibrating Noise to Sensitivity in Private Data Analysis
Dwork, C., McSherry, F., Nissim, K., & Smith, A. (2006) · Lecture Notes in Computer Science · doi:10.1007/11681878_14 · Cited by 7,272
Membership Inference Attacks Against Machine Learning Models
Shokri, R., Stronati, M., Song, C., & Shmatikov, V. (2017) · 2017 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp.2017.41 · Cited by 4,754 · cited 2×
“The threat is that an attacker learns by whom the training data was provided, known as membership inference [30, 44, 45, 57] and about whom it contains information, known as data extraction [9, 11, 29, 58, 68].”
BackgroundMethodNeural Architectures for Named Entity Recognition
Lample, G., Ballesteros, M., Subramanian, S., Kawakami, K., & Dyer, C. (2016) · Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies · doi:10.18653/v1/n16-1030 · Cited by 4,401 · cited 19×
“Moreover, (i) training NER requires domainspecific, labelled training data, (ii) NER models make errors (i.e. a subset of PII remain unrecognized), and (iii) existing, publicly available NER models only aim for deidentification but not anonymization [47].”
MethodInfluentialA survey of named entity recognition and classification
Nadeau, D., & Sekine, S. (2007) · Lingvisticae Investigationes · doi:10.1075/li.30.1.03nad · Cited by 2,507 · cited 1×
“When datasets are large, it is necessary to rely on Named Entity Recognition (NER) [47].”
MethodOur Data, Ourselves: Privacy Via Distributed Noise Generation
Dwork, C., Kenthapadi, K., McSherry, F., Mironov, I., & Naor, M. (2006) · Lecture Notes in Computer Science · doi:10.1007/11761679_29 · Cited by 1,835 · cited 2×
“PII scrubbing and Differential Privacy (DP) protect the privacy of training data at the cost of degrading model utility.”
Hierarchical Neural Story Generation
Fan, A., Lewis, M., & Dauphin, Y. (2018) · Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p18-1082 · Cited by 1,266 · cited 1×
“Text is generated iteratively by traversing the tree using greedy decoding, top-k sampling [18], or a beam search algorithm while conditioning the LM on all preceding tokens.”
Method
Cited by
Shen, X., Chen, Z., Backes, M., Shen, Y., & Zhang, Y. (2023) · arXiv · doi:10.48550/arxiv.2308.03825 · Cited by 861
Security and Privacy Challenges of Large Language Models: A Survey
Das, B. C., Amini, M. H., & Wu, Y. (2025) · ACM Computing Surveys · doi:10.1145/3712001 · Cited by 445
“It has been noted that paraphrasing would disrupt the sequence of injected data, such as injected instruction, and special character insertion.”
Shen, X., Chen, Z., Backes, M., Shen, Y., & Zhang, Y. (2024) · Proceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security · doi:10.1145/3658644.3670388 · Cited by 441
Large Language Models in Medicine: The Potentials and Pitfalls: A Narrative Review
Omiye, J. A., Gui, H., Rezaei, S. J., Zou, J., & Daneshjou, R. (2024) · Annals of Internal Medicine · doi:10.7326/m23-2772 · Cited by 361
Foundation metrics for evaluating effectiveness of healthcare conversations powered by generative AI
Abbasian, M., Khatibi, E., Azimi, I., Oniani, D., Shakeri Hossein Abad, Z., Thieme, A., Sriram, R., Yang, Z., Wang, Y., Lin, B., et al. (2024) · npj Digital Medicine · doi:10.1038/s41746-024-01074-z · Cited by 213
Metadata
- Authors
- Lukas, N., Salem, A., Sim, R., Tople, S., Wutschitz, L., & Zanella-Béguelin, S.
- Venue
- 2023 IEEE Symposium on Security and Privacy (SP)
- Publisher
- IEEE
- Published
- 2023-05
- Pages
- 346-363
- ISSN
- 1081-6011 (ISSN-L)
- OpenAlex
- W4385679821
- Cited by
- 154
- References
- 33
- FWCI
- 29.11
- Type
- conference paper
- Language
- en
- Open access
- green
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Analyzing Leakage of Personally Identifiable Information in Language Models. https://app.cochat.ai/papers/10.1109/sp46215.2023.10179300
- The paper
- Lukas, N., Salem, A., Sim, R., Tople, S., Wutschitz, L., & Zanella-Béguelin, S. (2023). Analyzing Leakage of Personally Identifiable Information in Language Models. 2023 IEEE Symposium on Security and Privacy (SP). https://doi.org/10.1109/sp46215.2023.10179300
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.