Dataset · reshabhs
Prompt injection examples for chatbots
reshabhs/SPML_Chatbot_Prompt_Injection
System prompts and user attacks for training LLM defense against hacking and injections.
What's inside
The corpus contains between 10 and 100 thousand tabular and text records in English. Inside are realistic system prompts for chatbots paired with labeled user queries attempting a prompt injection attack. Authors focused specifically on practical bot operation scenarios rather than just classical jailbreaks.
What you can do with it
- Training a classifier-filter to detect prompt injections before feeding to an LLM
- Testing chatbot system prompt resilience against malicious queries
- Fine-tuning guardrails models for corporate AI assistants
- Tasks
- Text classification
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 1.7K
- Likes
- 34
Models trained on it
- mradermacher/monitor-qwen35-08b-GGUFText classification 766
- dannyliv/agent-guard-modernbert-baseText classification 284
- mradermacher/monitor-qwen35-2b-GGUFText classification 146
- yzdvai/tiny-injection-guardText classification 97
- samratduttaofficial/WaterSheepZero shot classification 81
- qwalora/parallax-guard-xsmallText classification 49
- ITLL/WaterSheepZero shot classification 33
- dannyliv/agent-guard-deberta-pi-baseText classification 29
How to load it
from datasets import load_dataset
ds = load_dataset("reshabhs/SPML_Chatbot_Prompt_Injection")