AI Threats: Data Poisoning, Prompt Injection & Model Theft | ISACA AAIA Ep.31

Episode 31 of the ISACA AAIA Exam Prep Series catalogs the threats specifically targeting AI systems.

In this episode you'll learn:
• Training Data Leakage — sensitive data exposed through models
• Data Poisoning — tainted training inputs corrupting outcomes
• Model Poisoning — compromised weights and behaviors
• Model Theft & Inversion — extraction and reconstruction attacks
• Prompt Injection & Evasion — adversarial inputs at runtime
• Vendor AI Threats — risks inherited from third-party models
• AI Solution Disruption — availability and DoS-style attacks

If you don't know the threats, you can't audit the controls.

Part of the 55-episode ISACA AAIA Exam Prep Series.

0:00 AI threats · the three attack surfaces auditors must know
0:15 Why this matters · advise leadership and challenge vendors
0:34 Build, run, use · the three places AI gets attacked
0:55 Each surface in detail · development, runtime, through-use
1:43 Training data leakage · the kitchen analogy and data exfiltration
2:50 Data poisoning · four entry points an attacker can hit
3:18 RAG and the open-book exam · poisoning the reference library
3:52 Model poisoning · tampering with the math itself
4:20 Model theft · the chef analogy and reverse engineering
5:20 Prompt injection · the hypnotized guard and indirect attacks
6:04 Model evasion · sneaking past the sensors with misspelled keywords
6:23 Model inversion · attribute, membership, and reconstruction attacks
7:04 Vendor and third-party risk · the AI supply chain
7:26 Solution disruption · denial-of-service against AI systems
7:51 Recap · development threats, use threats, and shared threats
8:23 Keep going · RooCloud.com for the full AAIA exam prep series

#ISACA #AAIA #AIThreats #PromptInjection #AIAudit

Видео AI Threats: Data Poisoning, Prompt Injection & Model Theft | ISACA AAIA Ep.31 канала RooCloud