Wirewalk Instituteof AI
sec-230 Intermediate Waitlist

AI Data Security: Training Data, RAG Stores and Leakage

For security, data and ML engineers who need to protect the data that trains, grounds and flows through AI systems, from training corpora to vector stores to prompt logs.

The curriculum is being prepared and reviewed. Enrollment is not open yet.

About this course

AI systems multiply copies of sensitive data: training and fine-tuning sets, embeddings in vector databases, cached prompts and completions, evaluation sets and observability logs. Each copy needs an owner, an access model and an integrity story. This course maps that lifecycle and anchors it to the joint CISA, NSA and FBI information sheet AI Data Security: Best Practices for Securing Data Used to Train and Operate AI Systems (May 2025), which centers data supply chain risk, poisoned data and drift, together with the OWASP entries on data and model poisoning, sensitive information disclosure and vector and embedding weaknesses.

You will establish provenance with integrity hashes and signed datasets, screen curated and scraped corpora for poisoning, and scrub secrets and personal data before training. The retrieval module is the heart of the course: enforcing source-system permissions at query time, isolating tenants in vector databases, understanding what embeddings reveal, and hardening ingestion pipelines. You then address leakage through conversation history, caches and fine-tuned model memorization, design data loss prevention for AI traffic, log prompts without creating a breach target, and assess vendor retention, training-use and deletion terms. Formal privacy techniques such as differential privacy are covered in sec-330. The Wirewalk Institute certificate attests to completion and assessment of this course; it is not endorsed by OWASP, MITRE, NIST or any vendor named here.

What you will be able to do

  • Map every copy of data an AI system creates and assign an owner, access model and retention rule to each
  • Apply provenance, integrity hashing and signing to training and fine-tuning datasets
  • Identify poisoning risk in curated, scraped and user-contributed data and choose screening controls
  • Enforce source-system permissions in retrieval and isolate tenants in vector databases
  • Design DLP, redaction and canary controls for prompts, outputs and logs
  • Evaluate AI vendor data terms for retention, training use, residency and deletion

Syllabus

5 modules · 16 lessons · about 12 hours

Module 1: The AI Data Lifecycle

Where data lives in an AI system and the guidance that frames its protection.
3 lessons
  1. Mapping Data Through Collection, Training, Fine-Tuning, Retrieval and Logs
  2. CISA, NSA and FBI AI Data Security Guidance: Supply Chain, Poisoning and Drift
  3. Classifying Data for AI Use Cases

Module 2: Securing Training and Fine-Tuning Data

Integrity, provenance and hygiene for the data that shapes a model.
3 lessons
  1. Provenance, Integrity Hashes and Signed Datasets
  2. Screening Curated, Scraped and User-Contributed Data for Poisoning
  3. Removing Secrets and Personal Data Before Training

Module 3: Securing Retrieval and Vector Stores

Making RAG respect permissions and keep tenants apart.
4 lessons
  1. Permission-Aware Retrieval: Enforcing Source ACLs at Query Time
  2. Multi-Tenant Vector Databases: Namespaces, Indexes and Cluster Isolation
  3. What Embeddings Reveal: Inversion and Attribute Leakage
  4. Ingestion Pipelines as an Attack Surface

Module 4: Leakage Through Prompts, Outputs and Models

How sensitive data escapes in operation and how to catch it.
3 lessons
  1. Sensitive Data in Conversation History, Caches and Shared Chats
  2. Memorization and Extraction Risk in Fine-Tuned Models
  3. DLP for AI Traffic: Gateways, Redaction and Canary Tokens

Module 5: Logs, Retention and Third Parties

Observability without a breach target and realistic vendor controls.
3 lessons
  1. Logging Prompts and Completions Without Building a Breach Target
  2. Vendor Data Terms: Retention, Training Use, Residency and Zero Data Retention
  3. Deletion Requests and Unlearning: What Is Realistic

Exam and certificate

The exam unlocks once you have completed every lesson. It draws 20 questions from a larger bank, spread across the modules, with the answer options shuffled for each attempt. You have 45 minutes and need 80% to pass. After a failed attempt you can try again after 24 hours, up to 3 attempts in total. Feedback is given per module, never as the answer key.

Passing earns a certificate of completion and assessment from the Wirewalk Institute of AI with a unique ID that anyone can check at institute.wirewalk.com/verify. It is not an accredited qualification or a vendor certification. How certification works.

The Wirewalk Institute of AI is independent. This course teaches publicly documented products and is not affiliated with or endorsed by CISA, NSA, OWASP, or NIST. Product and company names are trademarks of their respective owners.

More in AI Security: Securing AI Systems

All 13 courses in this track
sec-001 Free intro Available

Using AI Safely at Work: The Essentials

Follow one fictional US distributor through four everyday AI mistakes (a guarantor's credit file uploaded to a personal account, a customer letter promising a fee waiver that does not exist, a vendor email with hidden instructions for the assistant, and a note-taker that sent a customer the internal margin floor) and practice the decisions that prevent or contain each one in three short Decision Labs.

1 h · 7 lessons Free intro
sec-101 Basic Available

Using AI Safely: Data Handling for Every Employee

A plain-language course on handling data safely with AI tools: what happens to what you type or upload, how to classify before you paste, how connected apps and agents change the picture, and how to report a mistake quickly.

6 h · 15 lessons See price at checkout
sec-201 Intermediate Available

AI Security Fundamentals: The OWASP Top 10 for LLMs

A practitioner course on securing applications that use large language models, organized around the 2026 OWASP Top 10 for LLM Applications. Every risk comes with a hypothetical scenario, the root cause, and concrete defenses in code and configuration.

14 h · 18 lessons See price at checkout