The AI Kill Chain
summary
Shows a concrete AI kill chain: a malicious PyPI library is planted, a resume white-on-white prompt injects instructions into an ATS, retrieval poisoning and agent memory persistence stores the instruction, and an autonomous agent abuses its email tool to add a weekly digest BCC to evilcorp.io, exfiltrating PII. Demo reproduces the chain end-to-end; persistence survives a system restart. Practical fixes are least-privilege agents, tool authorization, trust boundaries, explicit human approvals, and treating all external inputs as data not instructions.
AI-written summary · about
chapters (22)
- 0:00AI security threats intro
- 1:08Attack surface expansion
- 3:00Model to agent attacks
- 4:25OWASP agentic attacks
- 5:07AI failures and privileges
- 6:12Chaining AI attacks
- 7:57Real AI breach examples
- 9:03AI kill chain concept
- 10:28TalentFlow AI scenario
- 14:25TalentFlow AI app
- 15:32Resume context poisoning
- 17:16Recruiter weekly digest
- 18:32Malicious resume injection
- 19:39Agent memory poisoning
- 21:15Memory persistence impact
- 22:29Hidden text instructions
- 24:42Memory after restart
- 25:49Human in the loop defense
- 27:42Human approval actions
- 29:06Breaches and chain-breaking
- 30:07Data versus instructions
- 31:57Defending prompt injection
related talks
33mAgentic ProbLLMs: Exploiting AI Computer-Use and Coding Agents
36mAgentic AI Under Attack: Live Demos of Exploits Through Autonomy and Trust
51mPrompt Injection: When Hackers Befriend Your AI
53mHow to Break AI Systems (Before Someone Else Does)
31mAgentic AI Threats: How Multi-Agent Systems Get Exploited
30mThe Simple, Yet Lethal, Anatomy of an AI Attack
37mHow to Break AI Systems (Before Someone Else Does)
01h 00mHacking and Securing AI Systems: Playing With Fire and Controlling the Flare of AI