SQL Injection Jailbreak: a structural disaster of large language models
Paper
•
2411.01565
•
Published
You need to agree to share your contact information to access this dataset
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
You agree not to do anything illegal or unethical by applying the contents of the dataset and to use the dataset solely for research purposes.
Log in or Sign Up to review the conditions and access this dataset content.
WARNING: This dataset contains undesirable content generated by the LLMs, which does not represent the author's own wishes, and all experimental results are used for research purposes only!
Dataset Details
Note that due to a bug in the baseline evaluation code in this repository, some harmful scores do not match the actual scores. Please use the related experimental results with caution.
This is the dataset of experimental results for Sql Injection Jailbreak: a structural disaster of large language models containing results from the following experiments:
- Baseline Result: including experimental results from GCG, AutoDAN, ReNeLLM, DeepInception.
/other_baseline - SIJ experimental results: includeing the results of the experiments set by default in the paper.
/exp_result - Defense experimental results: including Self-Reminder, ICD, SafeDecoding, LLM-defend experiment results.
/baseline_defense_result - Self-Reminder-Key experimental results: including Self-Reminder-Key's experimental results.
defense_result - Normalized result: includes normalized results from SIJ and baseline attacks.
attack_norm_result
Github Repository
- Downloads last month
- 6