Keynote Panel: The Future of...
summary
Panel argues AI security needs hardware-to-cloud pipelines and shows concrete RISC‑V work: new extensions called Cove and coio to bind trusted sensors to confidential workloads, memory protection tables for isolation, and high assurance crypto key handles to prevent model exfiltration. Attestation, secure boot, SPDM, and external audits are recommended for runtime integrity. Risks highlighted: training‑data contamination, model cloning, and unsecured RAG deployments; cost and provenance of models motivate encryption, signatures, and verifiable delivery chains.
AI-written summary · about
chapters (14)
- 0:00Introduction to AI and security
- 1:45AI for security vs secure AI products
- 3:11Privacy risks in smart devices
- 6:22Building trust in AI systems
- 8:49Security in edge AI and IoT
- 10:32Behavior differences in AI security
- 13:46Protecting sensor data access
- 16:42Investment protection for AI models
- 20:14Data and model integrity risks
- 23:22Model watermarking and provenance
- 24:26Runtime code security checks
- 25:28AI as a security tool for defenders
- 26:33Call to action for secure AI
- 28:18Community collaboration on AI security
related talks
34mKeynote Panel: The Sovereign Stack: Nations, Enterprises & the Demand for Verifiable AI
14mKeynote: Back to the Future: Next-Generation Cloud Native Security
20mKeynote: Future-Proofing Confidential AI
04mKeynote: Welcome – Nelly Porter
15mKeynote: The Security Symphony
16mKeynote: Navigating Turbulent Waters of AI: Building a Secure and Safe Foundation
01mKeynote: Welcome Back
31mKeynote Panel: Gen AI: Opportunites & risks