Multi-step Jailbreaking Privacy Attacks on ChatGPT.
Haoran LiDadi GuoWei FanMingshi XuJie HuangFanpu MengYangqiu SongPublished in: EMNLP (Findings) (2023)
Keyphrases
- multi step
- security risks
- data publishing
- security mechanisms
- protection mechanisms
- lower bounding
- security threats
- personal information
- security issues
- biometric template
- single step
- privacy preserving
- cryptographic protocols
- k nearest neighbor
- personal data
- private information
- countermeasures
- privacy preservation
- privacy protection
- sensitive information
- watermarking scheme
- machine learning
- neural network
- differential privacy
- privacy issues
- access control
- knn
- pairwise
- training data
- privacy concerns
- nearest neighbor
- active learning
- learning process
- security measures