Get the app
extract 0 0 0

nandhak12/finguard-finance-injection-dataset

published by @nandhak12

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
524
Last 30 days
71
Likes
-
Size
5.3 MB
Created on HF
2026-03-27
Age
6 months ago

Description

FinGuard: Finance-Specific Prompt Injection Detection Dataset

Dataset Summary

FinGuard is the first open dataset for detecting prompt injection attacks

against agentic financial AI systems. It combines 6 public datasets with

synthetically generated finance-specific attack examples across 4 enterprise

agent types.

Dataset Structure

Split

Rows

SAFE

ATTACK

Train

10,699

5,375 (50.2%)

5,324 (49.8%)

Test

3,047

2,006 (65.8%)

1,041 (34.2%)… See the full description on the dataset page: https://huggingface.co/datasets/nandhak12/finguard-finance-injection-dataset.

Tags

task_categories:text-classificationlanguage:enlicense:apache-2.0size_categories:10K<n<100Kformat:csvmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:usprompt-injectionfinancellm-securityagentic-aijailbreakbanking

README current version from Hugging Face

license: apache-2.0
task_categories:

  • text-classification
    language:
  • en
    tags:
  • prompt-injection
  • finance
  • llm-security
  • agentic-ai
  • jailbreak
  • banking
    size_categories:
  • 10K<n<100K

FinGuard: Finance-Specific Prompt Injection Detection Dataset

Dataset Summary

FinGuard is the first open dataset for detecting prompt injection attacks
against agentic financial AI systems. It combines 6 public datasets with
synthetically generated finance-specific attack examples across 4 enterprise
agent types.

Dataset Structure

Split Rows SAFE ATTACK
Train 10,699 5,375 (50.2%) 5,324 (49.8%)
Test 3,047 2,006 (65.8%) 1,041 (34.2%)

Schema

Column Description
user_message User input sent to the financial agent
label SAFE or ATTACK
category Attack subcategory or benign type
agent_type Which agent received the message
available_tools Tools the agent has access to
source Origin dataset
split train or test

Agent Types

Agent Tools
banking_agent verify_user, check_balance, transfer_funds, manage_card, process_refund
fraud_detection_agent verify_user, execute_sql_query, flag_suspicious_account, freeze_account
investment_agent verify_user, execute_trade, get_portfolio_value, rebalance_portfolio
enterprise_finance_agent verify_user, execute_sql_query, transfer_funds, access_audit_logs

Finance-Specific Attack Categories (Novel)

Category Description Count
authorization_bypass Override transaction limits, skip auth 200
account_data_exfiltration Extract other users financial data 200
sql_injection_via_nlp SQL manipulation via natural language 200
financial_fraud_execution Unauthorized money movement 200
role_escalation Adopt admin/auditor persona 200
investment_manipulation Unauthorized trades, bypass risk limits 200

Usage

import pandas as pd
train = pd.read_csv("hf://datasets/nandhak12/finguard-finance-injection-dataset/train.csv")
test  = pd.read_csv("hf://datasets/nandhak12/finguard-finance-injection-dataset/test.csv")
print(train["label"].value_counts())

Data Sources

Source License
PolyAI/banking77 CC-BY 4.0
neuralchemy/Prompt-injection-dataset Apache 2.0
xTRam1/safe-guard-prompt-injection Apache 2.0
deepset/prompt-injections Apache 2.0
reshabhs/SPML_Chatbot_Prompt_Injection MIT
jackhhao/jailbreak-classification Apache 2.0
Synthetic (Claude API) Apache 2.0

Related Work

  • SPML: A DSL for Defending Language Models Against Prompt Attacks (Sharma et al., 2024)
  • Palo Alto Networks: Beyond Jailbreaks (2026)
  • CompFly AI: The Trust Control Plane for Autonomous Agents

README history 4 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-03-27Upload README.md with huggingface_huba836b123 KB
    Loading...
  2. 2026-03-27Upload test.csv with huggingface_hub537990331 B
    Loading...
  3. 2026-03-27Upload train.csv with huggingface_hubfffa89831 B
    Loading...
  4. 2026-03-27initial commitb55631f31 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-03-28[bot] Conversion to Parquetopen1 💬#1
    Loading...

Continue on Hugging Face →

Related datasets same stage · extract

Anthropic/hh-rlhf
2.1M dl 2442 likes 90.4 MB
Dataset Card for HH-RLHF Dataset Summary This repository provides access to two different kinds of data: Human ...
djain95/sae-jailbreaks-cache
468.6K dl 1 likes 30.6 GB
mosaicml/dolly_hhrlhf
309.5K dl 113 likes 23.7 MB
Dataset Card for "dolly_hhrlhf" This dataset is a combination of Databrick's dolly-15k dataset and a filtered subset of Anthrop...
Dahoas/full-hh-rlhf
266.3K dl 104 likes 129.8 MB
Dataset Card for "full-hh-rlhf" Anthropic's HH dataset reformatted into prompt, chosen, rejected samples.
mlabonne/harmful_behaviors
198.2K dl 169 likes 23 KB
mlabonne/harmless_alpaca
190.2K dl 51 likes 1.2 MB
walledai/AdvBench
176.3K dl 123 likes 38 KB
Dataset Card for AdvBench Paper: Universal and Transferable Adversarial Attacks on Aligned Language Models Data: AdvBench Datas...
JailbreakV-28K/JailBreakV-28k
116.4K dl 80 likes 288.8 MB
⛓‍💥 JailBreakV-28K: A Benchmark for Assessing the Robustness of MultiModal Large Language Models against Jailbreak Attacks 🌐 ...

Read further

Community best practices (extraction)
5 rules the field converged on for building the pair.
Building your own extraction pair
When the reference pair is not enough. Sizing, framing, and evaluation.
Datasets that shaped abliteration
The five corpora that did most of the work.
Non-English abliteration communities
How Chinese, Russian, Japanese, and Korean practitioners build extraction pairs.

Metadata

License
apache-2.0

Repository

FILE .gitattributes FILE README.md FILE test.csv FILE train.csv

File contents live on Hugging Face. Click a file or folder to open it there.

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python) 
from datasets import load_dataset
ds = load_dataset("nandhak12/finguard-finance-injection-dataset")
Raw snapshot (Python) 
from huggingface_hub import snapshot_download
snapshot_download(repo_id="nandhak12/finguard-finance-injection-dataset", repo_type="dataset")
git clone (requires git-lfs) 
git clone https://huggingface.co/datasets/nandhak12/finguard-finance-injection-dataset
Browse repository on Hugging Face ->
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Get the app Browse the catalog
Open in Abliteration