Human in the loop is not enough
summary
Claim: human review breaks at scale (10k–100k outputs/day) so security must move from manual controls to continuous AI-assisted pentesting, automated governance and runtime controls. Built an enterprise platform on third‑party LLMs and a demo using Ollama that shows document injection and secret exfiltration, then blocks those attacks with a layered guardrail (input/output/dialogue/retrieval) and a LamaGuard-style output filter. Recommend MITRE ATLAS/Maestro+STRIDE threat models, short‑lived keys, a single asset register and automated model quarantine / kill switch.
AI-written summary · about
chapters (10)
- 0:00AI trust and security challenges
- 3:36Tech evolution to autonomous AI
- 8:14Cybersecurity risks across AI eras
- 11:04Traditional controls fall short
- 12:29AI threat modeling
- 16:51AI governance frameworks
- 21:13AI guardrails for security
- 27:00AI-aware pentesting and cloud security
- 34:30AI defense pillars
- 38:30Human oversight insufficient
related talks
01h 01mIntroduction to AI Security
38mPhishing LLMs: Reeling in the Machine
30mNavigating the New AI Security Landscape
20mAI Red Teaming and Automation: Exploring Societal Risks in...
43mSecurity of GenAI & LLMs: A New Era of Ethical Hacking
18mCommit Secure AI: Embedding Security in Code, Pipelines, and Production
34mLLM Security 101 - An Introduction to AI Red Teaming
44mSecuring LLM-Powered Apps: Overcoming Security and Privacy Challenges