← AI-103 Course Home · MVCC · Dept. of Engineering & Technology · Dr. John L. Sands

🧭 Lesson Overview

Day 7 introduced OWASP LLM Top 10 as a practitioner threat reference and positioned it within the broader framework stack. Today you go deep on all ten items — not just what they are, but how each attack actually works, what a real compromise looks like, which defenses address it, and critically, which defenses sound good but don't work.

This lesson covers the 2025 edition, which introduced two genuinely new categories (System Prompt Leakage and Vector & Embedding Weaknesses) and renamed or expanded three others (Excessive Agency, expanded from the 2023 entry; Misinformation, refocused from 2023's Overreliance; and Unbounded Consumption, expanded from 2023's Model Denial of Service) — changes that reflect what practitioners are actually encountering in production LLM deployments, not theoretical future risks.

🎓 Anchoring Analogy: The Building Inspector's Code Book

A building inspector doesn't invent what to check on each inspection visit. They work from a code book — a ranked list of the failure modes that have actually caused buildings to collapse, catch fire, or flood, organized by frequency and severity. The OWASP LLM Top 10 is that code book for LLM application security.

Every item on the list represents a failure mode that practitioners are actually observing in deployed LLM applications. The ranking reflects community consensus on impact, exploitability, and prevalence. When a security team pen tests an LLM application, they work down this list — it's the agreed checklist that ensures they don't miss the failures that matter most in production.

Day 8 Learning Objectives

  1. Explain what the OWASP LLM Top 10 (2025) is, how it differs from the original OWASP Top 10, and what's new in the 2025 edition.
  2. For each of the 10 items: describe the vulnerability, explain the attack path, and identify the primary defense.
  3. Match real-world AI security incidents to specific OWASP LLM Top 10 categories.
  4. Distinguish between effective and ineffective defenses for each category.
  5. Map OWASP LLM items to ATLAS techniques and NIST AI 100-2 attack categories (cross-framework literacy).
  6. Build a prioritized defense matrix for a specified LLM application using the OWASP Top 10 as the organizing structure.

All 10 Items at a Glance — Click Any Card to Jump

🔓 Module 1 · Why OWASP LLM Top 10 Exists — and What's New in 2025

The traditional OWASP Top 10 for web applications (SQL injection, broken access control, XSS, and their relatives) was built around a fundamental assumption: the application executes code, and the code has bugs. Fix the bugs, secure the code, ship the application. LLMs violate this assumption completely — the "code" is the model's learned behavior, and you cannot inspect, verify, or patch it the way you patch a SQL query.

DimensionTraditional OWASP Top 10OWASP LLM Top 10 (2025)
Root causeCode bugs — deterministic, reproducible, patchableModel behavior — probabilistic, emergent, not directly patchable
Attack channelHTTP requests, SQL queries, form inputs, file uploadsNatural language prompts — instructions and data share the same channel
Testing methodologyFuzz inputs, check for injection, test access controlsAdversarial prompts, red-team role-play, systematic jailbreak testing
Defense philosophyInput validation, output encoding, parameterized queriesDefense in depth — model + application + infrastructure layers
Supply chainThird-party libraries with known CVEsFoundation model training data, model weights, fine-tuning datasets — no CVE system
Memory/contextApplications are stateless or have explicit session stateLLMs have context windows that accumulate across turns — injection can persist
Output behaviorOutputs are data — attacker manipulates inputs to get malicious outputsOutputs are natural language that downstream systems execute — model output is code

What Changed: 2023 → 2025 Edition

📋 2025 Edition Key Changes
Promoted: Sensitive Information Disclosure (#6 → #2), Supply Chain (#5 → #3 and expanded)
Demoted: Improper Output Handling (#2 → #5)
New in 2025: Excessive Agency (LLM06), System Prompt Leakage (LLM07), Vector & Embedding Weaknesses (LLM08), Misinformation (LLM09), Unbounded Consumption (LLM10)
Why these five? They reflect production realities: agentic AI deployments (LLM06), RAG & system-prompt risks (LLM07, LLM08), trust in AI outputs (LLM09), and pay-per-use cost attacks (LLM10).
OWASP LLM ItemATLAS TechniqueNIST AI 100-2 CategoryDeployment Type Most at Risk
LLM01 — Prompt InjectionAML.T0051, AML.T0054Evasion (GenAI)Any customer-facing LLM
LLM02 — Sensitive Info DisclosureAML.T0035, AML.T0024Privacy (training data extraction)Fine-tuned LLMs on private data
LLM03 — Supply ChainAML.T0048, AML.T0020Poisoning (supply chain)Any LLM using external models or data
LLM04 — Data & Model PoisoningAML.T0020, AML.T0019Poisoning (data)Fine-tuned and RAG-based LLMs
LLM05 — Improper Output HandlingAML.T0043Evasion → downstream injectionLLMs integrated with code/DB systems
LLM06 — Excessive AgencyAML.T0086, AML.T0110No direct NIST categoryAgentic AI with tool access
LLM07 — System Prompt LeakageAML.T0024Privacy (model extraction)Enterprise LLMs with custom prompts
LLM08 — Vector & Embedding WeaknessesAML.T0054, AML.T0020Poisoning (RAG)RAG-based applications
LLM09 — MisinformationNo direct ATLAS techniqueMisuse (GenAI)High-stakes information domains
LLM10 — Unbounded ConsumptionAML.T0029Availability attackPay-per-use API deployments
I
LLM01–LLM05 — The Foundational Five
⏱ ~60 min

🔴 Module 2 · LLM01–LLM05 — Flip Cards & Deep Dives

Flip each card to see the real-world attack scenario. Then expand each item below for the full technical detail, real incident, defenses, and what doesn't work.

II
LLM06–LLM10 — The 2025 Additions & Emerging Risks
⏱ ~50 min

🟣 Module 3 · LLM06–LLM10 — Flip Cards & Deep Dives

The five new additions in 2025 reflect real production deployment patterns — agents with tool access, RAG pipelines, system prompt governance, and resource cost attacks. Flip and expand each.

🛡️ Module 4 · Defense-in-Depth Architecture for LLM Applications

No single control addresses all 10 OWASP LLM items. Effective LLM security requires layering defenses across three distinct layers — each addressing specific items and filling gaps the other layers cannot cover.

Defense LayerWhat It CoversPrimary ControlsOWASP Items Addressed
Model LayerControls built into the model itself — safety training, RLHF, alignmentSafety fine-tuning, adversarial training, output classifiers on generation, RLHF guardrailsLLM01 (partial), LLM09 (partial)
Application LayerControls in the code around the model — input validation, output handling, access controlsInput sanitization, instruction-data separation, output schema validation, privilege reduction, system prompt hardening, content filters, rate limitingLLM01, LLM02, LLM05, LLM06, LLM07, LLM08, LLM10
Infrastructure LayerControls in the deployment environment — API security, monitoring, supply chain, data governanceVendor assessment, AIBOM, model provenance, API rate limiting, usage monitoring, DLP, incident response playbookLLM03, LLM04, LLM06, LLM10
⚠️ The Biggest Defense Misconception
"We added a system prompt instruction that says the model should not disclose sensitive information." This is a policy statement inside the model's context window — not a security control. The model cannot enforce this reliably against adversarial prompts. System prompts provide helpful guidance; they are not access controls. Every OWASP LLM mitigation that works lives at the application or infrastructure layer — not in the system prompt.
💡 The Security Guard Analogy

A system prompt instruction is like telling a brand-new employee: "Don't share confidential information." That's useful guidance for honest interactions. But it provides zero protection against a sophisticated social engineer. The actual controls — badge readers, locked filing cabinets, monitored exits, visitor logs — are in the physical infrastructure around the employee, not in the employee's verbal instructions.

LLM application security works the same way: the model is the employee, not the security system. The application layer provides the badge readers. The infrastructure layer provides the monitors. Defense-in-depth means not relying on the model to enforce its own security boundaries.

Lab Activities — Attack Lab, Case Match, Defense Builder
⏱ ~70 min

Activity A · OWASP Attack Scenario Lab

Match Each Scenario to the Correct OWASP LLM Item 0 / 10 correct

For each scenario, click the OWASP LLM item that best applies. Instant feedback explains the reasoning. Some scenarios are designed to be tricky — they may superficially resemble another item.

0 of 10 answered

🔍 Activity B · Real Incident Case Study Matcher

Match Real AI Security Incidents to OWASP LLM Categories 0 / 8 correct

Each entry describes a real or realistic documented AI security incident. Select the OWASP LLM Top 10 (2025) category that best classifies it. Click "Check All" when done.

🛡️ Activity C · Defense Builder — MVCC Enrollment Chatbot

Select the Effective Defenses for Each OWASP Item 0 / 10 scenarios

For each OWASP LLM item, a set of proposed controls is shown. Check every control that effectively addresses that specific risk. Some controls sound reasonable but don't actually work for that item — identify which.

📝 Assessment Artifact

📋 Day 8 Assessment Artifact

Open-Book Quiz: "Given This Scenario, Which OWASP LLM Category Applies and Why?"

The assessment artifact for Day 8 is the Day 8 Knowledge Check quiz below — completed open-book, submitted with explanations. For each question where you select an answer, write one sentence explaining why that category applies and one sentence identifying the primary defense. Submit your answers and explanations as a document by end of class.

Additionally: On a half-sheet, rank the OWASP LLM Top 10 items by priority for the MVCC Enrollment Chatbot (from highest to lowest) with a one-sentence justification for your top 3 choices. This ranking feeds into Day 11's secure architecture design.

Grading: Daily assignment (25% of course grade). Evaluated on accuracy of category identification, quality of defense reasoning, and appropriateness of the MVCC priority ranking.

Day 8 Knowledge Check

Day 8 Knowledge Check — 10 Questions Score: 0 / 10
Formative — this knowledge check does not affect your grade. Use it to self-assess before moving on.