The field of large language models is moving towards addressing pressing concerns of privacy and security. Researchers are exploring innovative solutions to protect user data and prevent malicious attacks on these models. One direction is the development of privacy-preserving frameworks that separate sensitive from non-sensitive data, allowing for secure processing of user interactions. Another area of focus is the detection and mitigation of adversarial attacks, with techniques such as prompt desensitization and reward neutralization showing promise. Additionally, there is a growing emphasis on securing the AI supply chain, including the detection of malicious configurations in model repositories. Noteworthy papers in this area include 'Preserving Privacy and Utility in LLM-Based Product Recommendations', which proposes a hybrid framework for privacy-preserving recommendations, and 'Fight Fire with Fire: Defending Against Malicious RL Fine-Tuning via Reward Neutralization', which introduces a defense framework against malicious reinforcement learning fine-tuning attacks.
Enhancing Privacy and Security in Large Language Models
Sources
Preserving Privacy and Utility in LLM-Based Product Recommendations
LLM Security: Vulnerabilities, Attacks, Defenses, and Countermeasures
A Rusty Link in the AI Supply Chain: Detecting Evil Configurations in Model Repositories
Anti-adversarial Learning: Desensitizing Prompts for Large Language Models
Helping Big Language Models Protect Themselves: An Enhanced Filtering and Summarization System
A Survey on Privacy Risks and Protection in Large Language Models
Advancing Email Spam Detection: Leveraging Zero-Shot Learning and Large Language Models
Unveiling the Landscape of LLM Deployment in the Wild: An Empirical Study
Avoid Recommending Out-of-Domain Items: Constrained Generative Recommendation with LLMs
Automatic Calibration for Membership Inference Attack on Large Language Models
A Comprehensive Analysis of Adversarial Attacks against Spam Filters
Fight Fire with Fire: Defending Against Malicious RL Fine-Tuning via Reward Neutralization
Personalized Risks and Regulatory Strategies of Large Language Models in Digital Advertising
Reliably Bounding False Positives: A Zero-Shot Machine-Generated Text Detection Framework via Multiscaled Conformal Prediction