Trust Is the Next Bottleneck: Why the Agentic Economy Needs Confidential
summary
Claims that current agents cannot cryptographically prove what code ran or whether data stayed inside a trustworthy boundary and gives hard figures — 750k agents compromised in one night, 195 million breached records, 150M poisoned MCP downloads. Shows an end-to-end payment-approval agent using confidential computing primitives (hardware attestation, measured execution) and proposes a six-layer attested-agent stack where layers 1–2 are production-ready but layers 3–6 (measured policy-as-code, attested MCP servers, portable evidence, cross‑cloud federation) need standard work and RFC 8693 hops must carry TE measurements.
AI-written summary · about
chapters (11)
- 0:00Agentic economy trust problem
- 1:02Current agent security limits
- 3:08Payments and hardware roots
- 5:35Hardware layer for agent identity
- 8:25Gaps in attested agent security
- 11:40Attested agent identity stack
- 13:25Agents' secure questions
- 15:33Hardware trust limits
- 16:56Market for attested identity
- 17:38Call to action
- 19:21Closing & Q&A
related talks
41mAgentic AI, confidential computing and attested TLS
40mDon’t Trust Opaque Clouds, Cryptographically Verify Instead!
40mIn AI We Trust? Securing the Future, One Agent at a Time
37mIt's All About Trust: Why Confidential Computing Matters
36mSecurity for AI and Multi-Party Collaboration with Confidential Computing (Oh, and...
35mChains of trust in Confidential Computing- 37mIt's All About Trust: Why Confidential Computing Matters
23mConfidential computing: from niche to mainstream