mlsanityData to 14 Oct 2026

InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy

arXiv 2507.02974 ↗First posted 30 Jun 2025cs.LG cs.CL cs.CRcited by 10

On the map: Retrieval-augmented generation → Privacy and security of RAG

Abstract (arXiv)

As major progress in LLM-based long-form text generation enables paradigms such as retrieval-augmented generation (RAG) and inference-time scaling, safely incorporating private information into the generation remains a critical open question. We present InvisibleInk, a highly scalable long-form text generation framework satisfying rigorous differential privacy guarantees with respect to the sensitive reference texts. It interprets sampling from the LLM's next-token-distribution as the exponential mechanism over the LLM logits with two innovations. First, we reduce the privacy cost by isolating and clipping only the sensitive information in the model logits (relative to the public logits). Second, we improve text quality by sampling without any privacy cost from a small superset of the top-k private tokens. Empirical evaluations demonstrate a consistent 8× (or more) reduction in computation cost over state-of-the-art baselines to generate long-form private text of the same utility across privacy levels. InvisibleInk is able to generate, for the first time, high-quality private long-form text at less than 4-8× times the computation cost of non-private generation, paving the way for its practical use. We open-source a pip-installable Python package (invink) for InvisibleInk at https://github.com/cerai-iitm/invisibleink.

Key facts. Cited by 10 papers in the map (data through 14 Oct 2026). Placed in Privacy and security of RAG, one of 15 directions in Retrieval-augmented generation. Builds on 71 of its 94 references that are in the map. First posted on arXiv 30 Jun 2025.

Builds on

71 of 94 references.

References · 71

  1. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2019Sentence-BERT: Sentence Embeddings using Siamese BERT-Networkscited by 9,170
  4. 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Modelscited by 3,211
  5. 2015Neural Machine Translation of Rare Words with Subword Unitscited by 2,918
  6. 2019The Curious Case of Neural Text Degenerationcited by 2,916
  7. 2025s1: Simple test-time scalingcited by 1,341
  8. 2018Hierarchical Neural Story Generationcited by 1,193
  9. 2021Show Your Work: Scratchpads for Intermediate Computation with Language Modelscited by 886
  10. 2017Renyi Differential Privacycited by 650
  11. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  12. 2024TinyLlama: An Open-Source Small Language Modelcited by 618
38 more
  1. 2021MAUVE: Measuring the Gap Between Neural Text and Human Text using Divergence Frontierscited by 470
  2. 2023Scalable Extraction of Training Data from (Production) Language Modelscited by 434
  3. 2016Concentrated Differential Privacy: Simplifications, Extensions, and Lower Boundscited by 346
  4. 2024Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Modelscited by 290
  5. 2023How to DP-fy ML: A Practical Guide to Machine Learning with Differential Privacycited by 153
  6. 2024Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputscited by 150
  7. 2024From Decoding to Meta-Generation: Inference-time Algorithms for Large Language Modelscited by 137
  8. 2022Truncation Sampling as Language Model Desmoothingcited by 123
  9. 2022Differentially Private Diffusion Modelscited by 119
  10. 2022Locally Typical Samplingcited by 111
  11. 2022Synthetic Text Generation with Differential Privacy: A Simple and Practical Recipecited by 108
  12. 2021Hyperparameter Tuning with Renyi Differential Privacycited by 104
  13. 2023Privacy-Preserving In-Context Learning with Differentially Private Few-Shot Generationcited by 96
  14. 2022The Text Anonymization Benchmark (TAB): A Dedicated Corpus and Evaluation Framework for Text Anonymizationcited by 95
  15. 2022Position: Considerations for Differentially Private Learning with Large-Scale Public Pretrainingcited by 88
  16. 2021Optimal Accounting of Differential Privacy via Characteristic Functioncited by 86
  17. 2019Hypothesis Testing Interpretations and Renyi Differential Privacycited by 85
  18. 2019Generative Models for Effective ML on Private, Decentralized Datasetscited by 85
  19. 2023Flocks of Stochastic Parrots: Differentially Private Prompt Learning for Large Language Modelscited by 83
  20. 2022Differentially Private Language Models for Secure Data Sharingcited by 68
  21. 2023Differentially Private Synthetic Data via Foundation Model APIs 1: Imagescited by 65
  22. 2023Harnessing large-language models to generate private synthetic textcited by 65
  23. 2023Locally Differentially Private Document Generation Using Zero Shot Promptingcited by 63
  24. 2021Iterative Methods for Private Synthetic Data: Unifying Framework and New Methodscited by 59
  25. 2023DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineercited by 57
  26. 2023Privacy-Preserving In-Context Learning for Large Language Modelscited by 50
  27. 2022Private Convex Optimization via Exponential Mechanismcited by 49
  28. 2022Composition of Differential Privacy & Privacy Amplification by Subsamplingcited by 45
  29. 2025Inference-Time Scaling for Complex Tasks: Where We Stand and What Lies Aheadcited by 44
  30. 2021TEM: High Utility Metric Differential Privacy on Textcited by 43
  31. 2023Explaining How Transformers Use Context to Build Predictionscited by 43
  32. 2021Leveraging Public Data for Practical Private Query Releasecited by 39
  33. 2022MAUVE Scores for Generative Models: Theory and Practicecited by 36
  34. 2023User Inference Attacks on Large Language Modelscited by 36
  35. 2022Private Synthetic Data for Multitask Learning and Marginal Queriescited by 35
  36. 2024Privacy-Preserving Instructions for Aligning Large Language Modelscited by 35
  37. 2022Differentially Private Decoding in Large Language Modelscited by 34
  38. 2023Closing the Curious Case of Neural Text Degenerationcited by 32

50 most cited of 71 All 71 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 1; 2026: 9.

10 papers.

  1. 2026Towards Secure Retrieval-Augmented Generation: A Comprehensive Review of Threats, Defenses and Benchmarkscited by 4
  2. 2025Private-RAG: Answering Multiple Queries with LLMs while Keeping Your Data Privatecited by 3
  3. 2026EPSVec: Efficient and Private Synthetic Data Generation via Dataset Vectorscited by 2
  4. 2026Securing Retrieval-Augmented Generation: A Taxonomy of Attacks, Defenses, and Future Directionscited by 2
  5. 2026Differentially Private Sampling from Distributions via Wasserstein Projectioncited by 0
  6. 2026Is External Database Protection Static in Retrieval-Augmented Generation? Rethinking Privacy Preservation under Dynamic Queriescited by 0
  7. 2026Auditing and Mitigating Privacy Leakage in Cloud-Edge Collaborative Decodingcited by 0
  8. 2026PrivBench: A Holistic and Modular Benchmarking Platform for Evaluating Text-to-Text Privatizationcited by 0
  9. 2026Only Pay What You Must Spend: On-Demand Privacy Budget Payment for Differentially Private RAGcited by 0
  10. 2026PrivCert: Certifying Statement Support under Differential Privacycited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG)
  2. 2024Privacy-Preserving Retrieval-Augmented Generation with Differential Privacy
  3. 2025SynBench: A Benchmark for Differentially Private Text Generation
  4. 2023InferDPT: Privacy-Preserving Inference for Closed-box Large Language Model
  5. 2024Private Synthetic Text Generation with Diffusion Models
  6. 2025Privacy-Aware Decoding: Mitigating Privacy Leakage of Large Language Models in Retrieval-Augmented Generation
  7. 2024Differentially Private Knowledge Distillation via Synthetic Text Generation
  8. 2024RemoteRAG: A Privacy-Preserving LLM Cloud RAG Service