When Models Remember and Reveal Training Data

When Models Remember and Reveal Training Data

Path Outline

Progress0%

1What Makes an LLM App Hackable 2How an LLM App Works Under the Hood 3Mapping Any LLM App With MORALS 4Meet the Frameworks That Classify AI Risks

5Prompt Injection When Users Rewrite the Rules 6Indirect Prompt Injection Hidden in Documents 7Jailbreaks That Talk Models Past Safety Rules 8How System Prompts and Hidden Context Leak 9Hidden Commands in Images and Audio 10How Sensitive Data Leaks Through AI Apps

11When Models Remember and Reveal Training Data

12Poisoning the Data a Model Learns From 13How Loading a Model File Can Run Malicious Code 14Fake Packages Invented by AI Coding Assistants 15Fooling and Stealing Models Through Their Inputs

16Treating AI Output as Untrusted Input 17How Document Search Pipelines Get Poisoned or Leak 18When Confident AI Answers Are Wrong 19When AI Agents Have Too Much Power 20Hidden Instructions in Agent Tools and Coding Rules 21Ten Ways AI Agents Go Wrong

22Layering Defenses Because No Single Fix Works 23Red Teaming Your AI App Before Attackers Do 24Governing AI Risk with NIST and the EU AI Act 25Threat Modeling a Support Copilot End to End 26Testing a Sandboxed Chatbot Safely and Next Steps

Hide outline

Home
Paths
LLM Security Fundamentals: AI as an Attack Surface
When Models Remember and Reveal Training Data