Slide 6 of 27
Part 1 · What Is It?Slide 6
Slide 6 · Why It Happens
Most sensitive info disclosure isn't malicious.
👤
The Well-Meaning Employee
Unintentional Disclosure
Pastes confidential code or meeting notes into a chatbot to save time, with no idea it might resurface later.
🛠️
The Over-Permissioned Integration
Configuration Failure
An AI tool is wired up with broader database or API access than the task actually needs.
🐛
The Infrastructure Bug
Application Failure
A caching or session-handling defect serves one user's data back to another.
🎯
The Deliberate Attacker
Targeted Attack
Crafts input specifically to extract data the system was supposed to protect.

Only the last one looks like a “hack.” The first three are just normal operation gone wrong — which is why prevention (Part 4) has to cover process and configuration, not just security filters.

← BackNext → Disclosure vs Prompt Leakage