Hardening AI/ML Systems - The Next Frontier of Cybersecurity
summary
Panel documents concrete risks and mitigations: adversarial ML includes data poisoning and input tampering, researchers describe a sponge DDoS variant, and real incidents include a $77 million Shanghai fraud against facial recognition and MathGPT’s API exposure that crashed servers. Recommended defenses are explicit: prefer simpler models when sufficient, enforce data lineage and monitoring, run AI red teams, adopt MITRE ATLAS and NIST guidance, maintain human-in-the-loop approval, and prepare for third‑party foundation model risk and rapid low‑sophistication exploitation.
AI-written summary · about
chapters (9)
- 0:00Introduction and opening remarks
- 4:02AI assurance frameworks overview
- 13:15Adversarial machine learning threats
- 20:36Industries at risk and parallels to cybersecurity
- 25:15Mitigation strategies for AML attacks
- 31:28Collaboration and resilience in AI security
- 35:05Defining success in AI hardening
- 38:16Contrarian takes and future tech
- 42:12Recommended resources and closing
related talks
44mAdvancing Cybersecurity with AI and Machine Learning: Opportunities and Challenges
49mThe Intersection of AI and Cybersecurity
37mFundamentals of AI safety and security
35mHacking AI Systems: How to (Still) Trick Artificial Intelligence- 44mAdvancing Cybersecurity with AI and Machine Learning: Opportunities and Challenges
49mGenAI and Cybersecurity: Safeguarding the Digital Age
12mSecuring AI-Driven Finance
46mA Short Introduction to Hacking Artificial Intelligence