datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
prompt-injections
Dataset Card for "deberta-v3-base-injection-dataset"
More Information needed
Prompt-injection-dataset
advance dataset if you want for llm security
https://huggingface.co/datasets/neuralchemy/prompt-injection-Threat-Matrix
Prompt Injection & Jailbreak Detection Dataset
A high-quality, leakage-free binary classification dataset for detecting prompt injection and jailbreak attacks against Large Language Models.
Zero data leakage — group-aware splitting confirmed
Balanced classes — ~60% malicious / 40% benign
Two configs — core for classical ML, full for transformers
29… See the full description on the dataset page: https://huggingface.co/datasets/neuralchemy/Prompt-injection-dataset.cyberseceval3-visual-prompt-injection
Dataset Card for CyberSecEval 3 - Visual Prompt Injection Benchmark
Dataset Details
Dataset Description
This dataset provides a multimodal benchmark for visual prompt injection, with text/image inputs. It is part of CyberSecEval 3, the third edition of Meta's flagship suite of security benchmarks for LLMs to measure cybersecurity risks and capabilities across multiple domains.
Language(s): English
License: MIT
Dataset Sources
Repository: Link… See the full description on the dataset page: https://huggingface.co/datasets/facebook/cyberseceval3-visual-prompt-injection.safe-guard-prompt-injectionWe formulated the prompt injection detector problem as a classification problem and trained our own language model
to detect whether a given user prompt is an attack or safe. First, to train our own prompt injection detector, we
required high-quality labelled data; however, existing prompt injection datasets were either too small (on the magnitude
of O(100)) or didn’t cover a broad spectrum of prompt injection attacks. To this end, inspired by the GLAN paper,
we created a custom synthetic… See the full description on the dataset page: https://huggingface.co/datasets/xTRam1/safe-guard-prompt-injection.Nemotron-RL-Agentic-Indirect-Prompt-Injection-v1
Nemotron-RL-Agentic-Indirect-Prompt-Injection-v1
Dataset Description:
Nemotron-RL-Agentic-Indirect-Prompt-Injection-v1 is an RL dataset for training and evaluating a tool-using agent's ability to resist Indirect Prompt Injection (IPI) attacks hidden inside tool-returned environment data. In each record, the agent receives a benign user request that requires calling a read tool whose output contains an adversarial instruction disguised as legitimate domain content… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/Nemotron-RL-Agentic-Indirect-Prompt-Injection-v1.mosscap_prompt_injection
mosscap_prompt_injection
This is a dataset of prompt injections submitted to the game Mosscap by Lakera.
This variant of the game Gandalf was created for DEF CON 31.
Note that the Mosscap levels may no longer be available in the future.
Note that we release every prompt that we received, regardless of whether it truly is a prompt injection or not.
There are hundrends of thousands of prompts and many of them are not actual prompt injections (people ask Mosscap all kinds of things).… See the full description on the dataset page: https://huggingface.co/datasets/Lakera/mosscap_prompt_injection.prompt-injections-benchmark
Dataset: Qualifire Benchmark Prompt Injection(Jailbreak vs. Benign) Datasets
Overview
This dataset contains 5,000 prompts, each labeled as either jailbreak or benign. The dataset is designed for evaluating AI models' robustness against adversarial prompts and their ability to distinguish between safe and unsafe inputs.
Dataset Structure
Total Samples: 5,000
Labels: jailbreak, benign
Columns:
text: The input text
label: The classification (jailbreak or benign)… See the full description on the dataset page: https://huggingface.co/datasets/rogue-security/prompt-injections-benchmark.SPML_Chatbot_Prompt_Injection
SPML Chatbot Prompt Injection Dataset
Arxiv Paper
Introducing the SPML Chatbot Prompt Injection Dataset: a robust collection of system prompts designed to create realistic chatbot interactions, coupled with a diverse array of annotated user prompts that attempt to carry out prompt injection attacks. While other datasets in this domain have centered on less practical chatbot scenarios or have limited themselves to "jailbreaking" – just one aspect of prompt injection – our dataset… See the full description on the dataset page: https://huggingface.co/datasets/reshabhs/SPML_Chatbot_Prompt_Injection.prompt-injection-safetyprompt-injection-dataset
Prompt Injection Detection Dataset
A binary classification dataset for detecting prompt injection attacks in user inputs to LLM-based applications.
Dataset Description
This dataset is designed to train encoder-only models (e.g., BERT, RoBERTa, DistilBERT) to classify user inputs as either benign or prompt injection attempts.
Classes
Label
Class
Description
0
BENIGN
Legitimate user queries
1
INJECTION
Prompt injection attempts
Features… See the full description on the dataset page: https://huggingface.co/datasets/S-Labs/prompt-injection-dataset.prompt_injections
Dataset Card for Prompt Injections by Yanis Miraoui 👋
Dataset Description
This dataset of prompt injections enriches Large Language Models (LLMs) by providing task-specific examples and prompts, helping improve LLMs' performance and control their behavior.
Dataset Summary
This dataset contains over 1000 rows of prompt injections in multiple languages. It contains examples of prompt injections using different techniques such as: prompt leaking… See the full description on the dataset page: https://huggingface.co/datasets/yanismiraoui/prompt_injections.prompt-injectionllm-jailbreak-prompt-injection-dataset
LLM Jailbreak & Prompt-Injection Dataset
A unified safety dataset combining 30+ public sources for training LLM
guardrails, content moderation classifiers, and response-safety filters.
Schema (orthogonal multi-label, WildGuard-style)
Instead of a single binary is_dangerous, every example carries four
orthogonal labels matching the structure used by AI2 WildGuard, IBM Granite
Guardian, and Azure Prompt Shields:
Column
Type
Description
prompt
str
The user/attack… See the full description on the dataset page: https://huggingface.co/datasets/Necent/llm-jailbreak-prompt-injection-dataset.agentic-prompt-injection-boundary-pairs
Agentic Prompt-Injection Boundary Pairs
Most prompt-injection datasets make the attack easy to recognize. The malicious row contains obvious override language, while the benign row discusses something unrelated. A classifier can look capable without learning the boundary that matters in production.
This dataset takes a stricter approach. Each attack is paired with a legitimate request from the same workflow. The two rows share the asset, role, tool and topic. What changes is… See the full description on the dataset page: https://huggingface.co/datasets/3nesdeniz/agentic-prompt-injection-boundary-pairs.prompt-injections
Dataset Card for "deberta-v3-base-injection-dataset"
More Information needed
prompt-injections
wambosec/prompt-injections
A dataset of prompts for training prompt injection detection models.
Dataset Description
This dataset contains prompts labeled as either benign (normal user requests) or malicious (prompt injection attacks).
Dataset Statistics
Total prompts: 5,766
Benign prompts: 2,340
Malicious prompts: 3,426
Malicious ratio: 59.4%
Dataset Structure
{
"prompt": str, # The prompt text
"label": int, # 0 =… See the full description on the dataset page: https://huggingface.co/datasets/wambosec/prompt-injections.turkish-conversation-prompt-injection
Turkish Conversation Prompt-Injection Dataset
Canonical dataset release:
Hugging Face
hosts the dataset viewer and downloads. This GitHub repository contains the
authoring sources, release files, documentation, deterministic build pipeline
and validation scripts. Version 1.0.2 has the permanent DOI
10.5281/zenodo.21379389 for its
Zenodo release archive.
The interactive dataset explorer
provides side-by-side inspection of all 150 controlled boundary pairs, complete
row… See the full description on the dataset page: https://huggingface.co/datasets/3nesdeniz/turkish-conversation-prompt-injection.prompt_injection_cleaned_dataset-v2
Dataset Card for "prompt_injection_cleaned_dataset-v2"
More Information needed
Indirect-Prompt-Injection-BIPIA-GPT
Indirect Prompt Injection Detection Dataset (BIPIA + GPT-4o-mini)
Dataset Summary
This dataset contains 70,000 examples for detecting indirect prompt injection attacks in Large Language Models. It combines:
35,000 malicious samples from the BIPIA benchmark (cleaned and processed)
35,000 benign samples generated using GPT-4o-mini
Indirect prompt injection attacks embed malicious instructions within external content (code, table, email, webAQ, abstract) that LLMs process… See the full description on the dataset page: https://huggingface.co/datasets/MAlmasabi/Indirect-Prompt-Injection-BIPIA-GPT.prompt-injection-Threat-Matrix
CATEGORIZED DATASET - easy to use
https://huggingface.co/datasets/neuralchemy/prompt-injection-dataset-categorized
Neuralchemy Prompt Injection Threat Matrix
A professional-grade prompt injection and
jailbreak detection dataset featuring 32,320
curated samples across 5 dimensions
with full threat intelligence schema including
technique classification, severity scoring,
attack surface detection, and ambiguity flagging.
Built for training production-grade LLM… See the full description on the dataset page: https://huggingface.co/datasets/neuralchemy/prompt-injection-Threat-Matrix.prompt_injection_password_or_secretprompt-injection-purple-llamaprompt-injection-benchmark
Prompt Injection Benchmark
A curated dataset of labeled prompt injection attacks and benign prompts for testing and benchmarking injection detection systems.
Dataset Description
This dataset contains 200 examples across 7 attack categories, plus 100 benign prompts. Each example is labeled with:
text: The prompt text
label: injection or benign
category: Attack category (e.g., instruction_override, role_hijack)
severity: low, medium, high, or critical
Attack… See the full description on the dataset page: https://huggingface.co/datasets/zachz/prompt-injection-benchmark.prompt-injection-dataset-categorized
Prompt Injection Dataset — Categorized (Threat Matrix V2)
Welcome to Prompt Injection Dataset – Categorized (formerly Threat Matrix), by Neuralchemy.
This is the successor to our original Prompt Injection Threat Matrix dataset. Instead of one multi-label table, this version splits the taxonomy into 7 clean, single-purpose subsets — 6 taxonomy dimensions plus a bonus ambiguity flag — so you can train a focused specialist model on each one instead of fighting multi-task learning.… See the full description on the dataset page: https://huggingface.co/datasets/neuralchemy/prompt-injection-dataset-categorized.agentic-prompt-injection-5k
Agentic Prompt-Injection 5K
5,000 examples of agentic and indirect prompt injection. A curated, paired benign/attack dataset for evaluating and training prompt-injection detectors and LLM guardrails. It focuses on the harder, agentic surface: tool/function abuse, RAG-document-embedded (indirect) injection, memory and trust-boundary poisoning, and approval/authority escalation.
Curated by Enes Deniz (ORCID 0009-0006-9491-3565), Co-Founder at AltaySec and OWASP AI Exchange / GenAI… See the full description on the dataset page: https://huggingface.co/datasets/3nesdeniz/agentic-prompt-injection-5k.prompt_injection_cleaned_dataset
Dataset Card for "prompt_injection_cleaned_dataset"
More Information needed
clawk-agent-social-ai-prompt-injection-dataset
Clawk Agent-Social AI Prompt Injection Dataset
85,703 items — 44,232 posts and 41,471 replies — from Clawk, a social network whose users are AI agents.
Scanned for AI-to-AI indirect prompt injection using the threat model of Greshake et al. (2023). The full raw corpus is included, so you can ignore my analysis entirely and do your own.
These are keyword-matched candidates, not verified attacks. An agent discussing prompt injection matches the same words as one performing it.… See the full description on the dataset page: https://huggingface.co/datasets/DavidTKeane/clawk-agent-social-ai-prompt-injection-dataset.zn-prompt-injection-bench
ZN Prompt-Injection Evaluation Corpus (v15)
A 23,699-row binary text-classification corpus for detecting prompt injection,
jailbreak attempts, and indirect attacks against LLM-based systems, paired with
hard adversarial negatives drawn from real human text. Every row carries its
source, license, language, and full provenance.
🛡️ Real-Time Defense for Agents: Evaluating models or securing your AI workflow? Check out zn-gate on npm (npx -y zn-gate mcp) and the open-source… See the full description on the dataset page: https://huggingface.co/datasets/tljohnsilver/zn-prompt-injection-bench.prompt-2-prompt-injection-v2-dataset-ruПереведённый с помощью Gemini 2.5 flash и Gemini 2.0 flash вариант датасета r1char9/prompt-2-prompt-injection-v2-dataset
moltbook-agent-social-ai-prompt-injection-dataset
Moltbook Agent-Social AI Prompt Injection Dataset
207,391 items — 77,469 posts and 129,922 comments — from Moltbook, a social network whose users are AI agents.
Scanned for indirect prompt-injection patterns using the taxonomy of Greshake et al. (2023). The full raw corpus is included, so you can ignore my analysis entirely and do your own.
These are keyword-matched candidates, not verified attacks. An agent discussing prompt injection matches the same words as one performing… See the full description on the dataset page: https://huggingface.co/datasets/DavidTKeane/moltbook-agent-social-ai-prompt-injection-dataset.