dataset: zivis-sim-fin
tags:

• synthetic
• financial
• security
• red-teaming
• langchain
• llm
  license: apache-2.0
  annotations_creators: [machine-generated]
  language: [en]
  pretty_name: ZIVIS-SIM Financial Red Team Dataset
  description: >
  A synthetic dataset of financial client profile documents created by the Zivis team (zivis.ai)
  for red teaming, LLM attack simulation, and sensitive information retrieval testing in AI pipelines.

🛡️ ZIVIS-SIM Financial Red Team Dataset

This dataset contains synthetically generated financial profile documents designed for security research, AI red teaming, and simulation of sensitive data exposure scenarios.

Created by the team at Zivis, this dataset is part of ongoing efforts to test and demonstrate real-world LLM risks in controlled environments.

📄 Dataset Contents

Each document simulates an internal financial case file at a fictional bank ("ZBank") and may include:

• Full name, SSN, DOB, address, phone, email
• Employment and income details
• Bank account numbers and balances
• Loan types and amounts
• Insurance policy information
• Routing/IBAN/SWIFT codes
• Credit scores
• Natural-language summary of the client’s financial posture

⚠️ All data is synthetic, generated using Faker and OpenAI GPT-4, and intended strictly for safe testing environments.

🧪 Use Cases

This dataset was created for use in the zivis-sim project — an open-source simulation environment for exploring AI vulnerabilities, particularly in:

• Retrieval-Augmented Generation (RAG)
• Vector search pipelines
• SSE (Server-Sent Events) streaming attack surfaces

You can use this dataset to:

• Test embedding/vector search leakage (e.g. vector inversion)
• Simulate SSE stream sniffing or hijacking
• Practice red teaming on synthetic PII-rich documents
• Benchmark document security workflows in LLM-based systems

⚠️ Disclaimer

Although this dataset is entirely fake and synthetically generated, it closely mimics real sensitive data.

Do not deploy this data in production environments or expose it publicly without proper safeguards.

It is intended solely for:

• Security research
• Simulation of privacy attacks
• Educational demonstrations of LLM vulnerabilities

🔗 Related Project

• 🔬 zivis-sim: Core simulation framework for adversarial testing of AI pipelines.