Keynote Panel Discussion: Three Forces Making Confidential AI A Mandate
summary
Three forces make Confidential AI mandatory: Mythos‑class models make a malicious agent inside your environment probable, current agent architectures leak secrets via context windows, RAG and MCP calls, and regulators now demand runtime, hardware‑signed proofs — notably EU AI Act Article 12 Aug 2 2026. Concrete responses include running agents on confidential computing (Department of Defense uses it for autonomous drones), instrumenting MCP servers as the “hands” with logs and audits, and scaling oversight with agent‑based code review — seen in Claude Tag and an observed 8x code per engineer productivity multiplier.
AI-written summary · about
chapters (11)
- 0:00Confidential AI urgency
- 1:06AI agents privacy and security
- 2:52Confidential computing necessity
- 7:47Non-deterministic AI challenges
- 11:16AI liability and compliance
- 13:23Agent architectures vulnerabilities
- 19:13Sensitive data handling
- 21:38AI agents observability
- 25:51Regulatory standards
- 28:20Industry readiness for AI regulation
- 31:13Closing remarks and Q&A
related talks
01h 11mKeynote - Securing AI: Building Trust in the Era of AI
41mAgentic AI, confidential computing and attested TLS
19mKeynote: Towards Building Safe and Secure Agentic AI
19mHow Can We Have Private Access to Intelligence?
30mKeynote Panel: The Future of...
40mIn AI We Trust? Securing the Future, One Agent at a Time
37mConfidential AI: Enabling Trustworthy Ecosystems in the Age of Agentic Intelligence - Session
35mAI and IoT security - The road ahead. Expert Panel