Slide 27 of 27
CompleteSlide 27
Complete
LLM02:2025 Complete
You've covered all 5 parts, read the real incidents, and tested yourself.
What sensitive information disclosure is — in plain English and in OWASP's definition
The seven categories of sensitive data, including the proprietary-algorithm category teams forget
4 disclosure types — each anchored to a real, confirmed incident
Disclosure vs. system prompt leakage (LLM02 vs LLM07) — what they share and where they differ
All 3 OWASP attack scenarios, each grounded in a real-world example
The OpenAI Redis bug (March 2023) — unintentional cross-user data exposure
Proof Pudding (CVE-2019-20634) — model extraction against a real spam filter
Samsung's 20-day ChatGPT leak (April 2023) and the company-wide ban that followed
The ChatGPT “poem” training data extraction attack (Nasr et al., 2023)
Slack AI's indirect prompt injection exfiltration (PromptArmor, 2024)
All 6 mitigation categories — what OWASP says, how real incidents made the gap worse, how to do it right, how to validate
Review from beginning Next: LLM03 →
📄 View all sources for this lesson