SafetyPage

Government · Government · NIST

NIST SP 800-239 initial public draft, AI Data Center Security Analysis: A High-Performance Computing (HPC) Driven Approach

Jul 2026

ReadSource at nvlpubs.nist.gov ↗Read herePublic domain (US government work)

Outline · 32 pages

  1. 1. Introductionp. 9
  2. 1.1. Zone-Based HPC Architecturep. 9
  3. 2. Reference Architecture and Differencesp. 12
  4. 2.1. AI Data Center Reference Architecturep. 12
  5. 2.2. Computing Zone Hardware Infrastructure Differencep. 14
  6. 2.2.1. Accelerator-Dense Infrastructurep. 14
  7. 2.2.2. Diverse AI Acceleratorsp. 14
  8. 2.2.3. Reduced-Precision Arithmeticp. 14
  9. 2.2.4. Use of High-Bandwidth Memoryp. 14
  10. 2.2.5. AI-Optimized High-Speed Networkingp. 15
  11. 2.3. Computing Zone Software Stack Differencep. 15
  12. 2.3.1. System Software and Deep Learning SDKp. 15
  13. 2.3.2. Deep Learning Librariesp. 15
  14. 2.3.3. Resource Management and Orchestrationp. 16
  15. 2.4. Data Storage Zone Differencep. 16
  16. 2.4.1. Use of Object Storage as the Backbonep. 16
  17. 2.4.2. Layered Data Storage Systemp. 16
  18. 2.5. Access Zone Differencep. 17
  19. 2.6. Management Zone Differencep. 17
  20. 2.7. Open Models and Datasetsp. 18
  21. 3. AI Data Center Security Threats and Challengesp. 19
  22. 3.1. AI Data Center Scalability Challengep. 19
  23. 3.2. Data Quality, Provenance, and Security Challengesp. 19
  24. 3.3. Regulatory and Compliance Challenges and Logging, Monitoring, and Audit-Ready Challengesp. 19
  25. 3.4. Virtual Environment Threatsp. 20
  26. 3.5. Denial-of-Service (DoS) Threatsp. 20
  27. 3.6. Prompt-Based Exploitation Threatsp. 21
  28. 3.7. Supply Chain Threatsp. 22
  29. 3.8. Multi-Tenant Threats and the Three-Way “Trust Dilemma”p. 22
  30. 3.9. Insider Threatsp. 23
  31. 3.10. Silent Data Corruption (SDC) Threatp. 23
  32. 3.11. Expanding and Evolving Attack Surface Threats and AI-Powered Attacksp. 23
  33. 3.12. Multiple Data Center Training Threatp. 23
  34. 3.13. Firmware Integrity Threatp. 24
  35. 3.13.1. Baseboard Management Controller (BMC) Threatsp. 24
  36. 3.13.2. Unified Extensible Firmware Interface (UEFI) Threatp. 24
  37. 3.14. AI Accelerator Vulnerabilityp. 25
  38. 4. AI Data Center Security Posture and Possible Solutionsp. 26
  39. 4.1. Strengthening the Access Zone Protectionp. 26
  40. 4.2. Providing Strong Monitoring, Logging, Security, and Compliance Oversight Capabilitiesp. 26
  41. 4.3. Zero-Trust vs. Nurturing Trustp. 27
  42. 4.4. Enabling Human-In-The-Loop Oversight and Governance Capabilityp. 27
  43. 4.5. Developing a Robust and Fault-Tolerant Workflowp. 28
  44. 4.6. Security-by-Design and Verification-by-Design Approach to AI Data Centerp. 28
  45. 5. Conclusionsp. 29
  46. Referencesp. 30