Creating an AI Security and Incident Response Team
summary
Carnegie Mellon CERT builds an AI Security and Incident Response Team, soft-launching in August and formally launching in October to triage reports from prompt exploits through hardware leakage and model serialization RCEs. The team adapts a CSIRT framework, adds an ML practitioner, uses SSVC for prioritization, struggles with reproducibility and test/verify for black‑box models, and coordinates disclosures and advisories via AISIRT launch: soft August, formal October, report to www.kb.cert.org, augment CSIRT with ML practitioner.
AI-written summary · about
chapters (11)
- 0:00AI Security Incident Response intro
- 1:23Need for AI Security team
- 3:36Real-world AI incidents
- 6:49AISIRT in vulnerability management
- 9:45AISIRT framework & root cause
- 13:02Situational awareness & best practices
- 14:28Lessons from first six months
- 19:31Regulatory & risk for AI
- 23:28Questions for AI adoption
- 26:37Contact & next steps
- 29:53Q&A on AI incident response
related talks
44mWhat Security Teams Need to Know and Do
26mImproving Cybersecurity Posture with an AI Security Assistant
01h 05mWhat It Takes to Lead America’s Vulnerability Management Team
14mEnhancing Enterprise Security with AI
40mApplying Artificial Intelligence to the Incident Response Function
28mFrom 2017 to 2021: Integration of an Operational Situation Awareness Team to a CSIRT...
01h 01mAI and the SOC Engineer: The Evolving Human Element in Security
32mAI/Machine Learning: What is Actually Working in Cybersecurity?