Instruction Backdoor Attacks Against Customized LLMs.
Rui ZhangHongwei LiRui WenWenbo JiangYuan ZhangMichael BackesYun ShenYang ZhangPublished in: USENIX Security Symposium (2024)
Keyphrases
- countermeasures
- instructional design
- learning disabled students
- multimedia
- malicious attacks
- watermarking scheme
- computer security
- security threats
- instruction set
- cryptographic protocols
- data corruption
- denial of service attacks
- traffic analysis
- intelligent tutor
- neural network
- learning experience
- information systems