Privacy-Preserving Prompt Tuning for Large Language Model

Symbol	🌟	⬜️	⬛️
Description	Inspiration	White-box method	Black-box method

Attacker Methodology by Stages

Prompt Injection Attacks (PIA)

Paper	Year	Source	Attack Prompt Type	Tasks
⬛️Effective Prompt Extraction from Language Models	2024.02		Instruction Prompt	Information Extration
⬛️Prompt Stealing Attacks Against Large Language Models	2024.02		Role-Based Prompt, Direct Prompt, In-Context Prompt	Q&A
🌟⬛️TrojLLM: A Black-box Trojan Prompt Attack on Large Language Models	(NeurIPS, 2023)		Instruction prompt	Classification
🌟⬛️Ignore Previous Prompt : Attack Techniques For Language Models	(NeurIPS, 2022)		Instruction prompt, In-context learning
🌟⬜️BadPrompt: Backdoor Attacks on Continuous Prompts	(NeurIPS, 2022)		Instruction prompt
⬜️PromptAttack: Prompt-based Attack for Language Models via Gradient Search	(NLPCC, 2022)		Instruction prompt

Membership Inference Attacks (MIA)

Paper	Year	Source
🌟⬛️Do Membership Inference AttacksWork on Large Language Models?	2024.02
⬜️Language Model Inversion	2023.11
⬛️Assessing Privacy Risks in Language Models: A Case Study on Summarization Tasks	2023.10
⬛️Beyond Memorization: Violating Privacy Via Inference with Large Language Models	2023.10
⬜️Extracting Training Data from Large Language Models	(USENIX Security, 2021)

Protector Methodology

Differential Privacy (DP)

Paper	Year	Source	Tasks	Defense
🌟⬛️Privacy-Preserving In-Context Learning with Differentially Private Few-Shot Generation	(ICLR, 2024)		Classification, Information Extraction
🌟⬛️DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineer	(ICLR, 2024)		Sentiment Classification
🌟⬛️Privacy-Preserving In-Context Learning For Large Language Models	(ICLR, 2024)		Classification, Document Q&A, Dialog Summarization
⬛️On the Privacy Risk of In-context Learning	(TrustNLP, 2024))		Classification, Generation	MIA
⬛️A Customized Text Sanitization Mechanism with Differential Privacy	(ACL, 2023)		Classification, Generation
🌟⬛️⬜️Flocks of Stochastic Parrots: Differentially Private Prompt Learning for Large Language Models	(NeurIPS, 2023)		Classification
🌟⬛️Locally Differentially Private Document Generation Using Zero Shot Prompting	(EMNLP, 2023)		Text Classification
⬜️DP-forward: Fine-tuning and inference on language models with differential privacy in forward pass	(SIGSAC, 2023)		Classification
⬛️InferDPT: Privacy-preserving Inference for Black-box Large Language Models	2023.12		Classification, Generation
⬜️Privacy-Preserving Prompt Tuning for Large Language Model Services	2023.05		Sentiment Classification, Document Q&A
⬛️Differential Privacy for Text Analytics via Natural Text Sanitization	(ACL-IJCNLP, 2021)		Classification

Secure Multi-Party Computing (SMPC)

Paper	Year	Source	Tasks	Defence
⬜️Ciphergpt: Secure two-party gpt inference	(Crypto, 2024)		Classification
⬜️SecFormer: Towards Fast and Accurate Privacy-Preserving Inference for Large Language Models	2024.01		Classification, Semantic Similarity, Linguistic Acceptability	Model Inside
⬜️LLMs Can Understand Encrypted Prompt: Towards Privacy-Computing Friendly Transformers	2023.05		Classification	Model Inside

Anonymization Techniques

Paper	Year	Source	Tasks	Keywords
⬛️SEmojiCrypt: Prompt Encryption for Secure Communication with Large Language Models	2024.02		Classification	Emoji
⬛️ProPILE: Probing Privacy Leakage in Large Language Models	(NeurIPS, 2023)		PII
⬛️Recovering from Privacy-Preserving Masking with Large Language Models	2023.12			[MASK]
⬛️Hide and Seek (HaS): A Lightweight Framework for Prompt Privacy Protection	2023.09		PII

Other Methods

Paper	Year	Source	Tasks	Method Keyword
⬜️Privatelora For Efficient Privacy Preserving LLM	(CoRR, 2023)			LoRA
⬜️TextObfuscator: Making Pre-trained Language Model a Privacy Protector via Obfuscating Word Representations	(ACL, 2023)		Classification

Related Survey

Paper	Year	Source
On Protecting the Data Privacy of Large Language Models (LLMs): A Survey	2024.03
Can LLMs Keep a Secret? Testing Privacy Implications of Language Models via Contextual Integrity Theory	(ICLR, 2024)
A Survey on Large Language Model (LLM) Security and Privacy: The Good, the Bad, and the Ugly	2023.12
Privacy in Large Language Models: Attacks, Defenses and Future Directions	2023.10
Not what you’ve signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection	2023.05
Privacy-Preserving Large Language Models (PPLLMs)	2023.01

Fine-tuning

Paper	Year	Source
SentinelLMs: Encrypted Input Adaptation and Fine-tuning of Language Models for Private and Secure Inference	(AAAI, 2024)