Episode 19: AI Data Poisoning: How Bad Actors Corrupt Machine Learning Systems for Under $60

TLP - The Digital Forensics Podcast

Send us a text

Clint Marsden breaks down a critical cybersecurity report from intelligence agencies including the CSA, NSA, and FBI about the growing threat of AI data poisoning. Learn how malicious actors can hijack AI systems for as little as $60, turning machine learning models against their intended purpose by corrupting training data.

Clint explains the technical concept of data poisoning in accessible terms, comparing it to teaching a child the wrong labels for objects. He walks through the six-stage framework where AI systems become vulnerable, from initial design to production deployment, and covers the ten security recommendations intelligence agencies are now promoting to defend against these attacks.

The episode explores real-world examples of AI systems gone wrong, from shopping bots buying drugs on the dark web to coordinated attacks by online communities. You'll discover practical mitigation strategies including cryptographic verification, secure data storage, anomaly detection, and the importance of "human in the loop" safeguards.

Whether you're a cybersecurity professional, AI developer, or simply curious about emerging digital threats, this episode provides essential insights into protecting AI systems from manipulation and understanding why data integrity has become a national security concern.

Key Topics Covered:

• Split view poisoning and expired domain attacks
• Data sanitization and anomaly detection techniques
• Zero trust principles for AI infrastructure
• The role of adversarial machine learning in cybersecurity
• Why defenders must learn AI as quickly as attackers
  The PDF from CISA etc al: https://www.ic3.gov/CSA/2025/250522.pdf