Hardware At The Edge Reliability, Ai Infrastructure And Defensive Design
summary
Argues that observability is the most underweighted defensive pillar, cites 66% of Meta training job interruptions are hardware failures, and warns silent data corruption can cost tens or hundreds of millions. Describes a six‑pillar framework and concrete fixes like capacitor orientation and heavier modular line cards that warped at 70 lb. Reports experiments feeding schematics, firmware and specs into Anthropic Mythos where Mythos produced exploit chains and suggested FPGA-based injections, prompting faster quarterly releases and earlier automated testing.
AI-written summary · about
chapters (16)
- 0:00Introduction and speaker background
- 2:20Hardware reliability at critical moment
- 3:22Challenges from hyperscalers and AI growth
- 5:05Defensive design framework pillars
- 7:52Simple but high-impact defensive designs
- 9:40Silliest defensive design that mattered
- 11:49AI infrastructure and software-hardware boundary
- 15:42Hardware testing lagging behind software
- 19:11Collapse of hardware-software security separation
- 20:37Why Cisco used Mythos on hardware
- 23:46AI for discovery vs remediation
- 25:09Defenders' advantage with AI tooling
- 27:19Impact on legacy and soft-metal products
- 29:43Shift left methodology in hardware
- 32:30Need for hardware transparency
- 35:19Blind spots in hardware security
related talks
36mFoundational Design Principles and Patterns for Building Robust & Adaptive(…)
20mThe AI bugpocalypse is here. Now what?
36mSurviving the Vulnpocalypse: Open Source and Supply Chain Security in a Post Mythos World.
20mO11y in the Evolution of AI-Era Networking Infrastructure
23mOpen, Intelligent, and Secure: The Next Industrial Edge
21mhow to reconcile high-performance hosting and innovation projects with IT security
25mYour Software Supply Chain is Only as Secure as its Weakest Link
44mFireside chat