Skip to main content

Confidential Computing for Enclava

Overview​

Enclava uses confidential computing to protect sensitive data during AI processing. This technology ensures your data remains encrypted even while being processed by AI models.

Key Concepts​

Trusted Execution Environments (TEEs): Hardware-secured areas where data is processed in isolation. Even cloud providers and system administrators cannot access data inside a TEE.

Remote Attestation: Cryptographic proof that your workload is running in a genuine, unmodified TEE before processing begins.

End-to-End Encryption: Data remains encrypted from client to TEE, during processing, and back to client.

Security Guarantees​

What TEEs Protect Against​

  • Cloud Provider Access: Administrators cannot view your data
  • Host OS Compromise: Malicious software on the host cannot access TEE memory
  • Physical Attacks: Hardware-level memory encryption prevents physical extraction
  • Co-tenant Threats: Other workloads on the same hardware cannot access your data

Known Limitations​

  • Some side-channel attacks remain partially mitigated
  • Performance overhead of 3-5% for encryption operations
  • Requires trust in hardware manufacturer's implementation

Technologies Used by Enclava​

NVIDIA H100 TEEs​

  • Hardware-accelerated AI inference with GPU memory encryption
  • Multi-Instance GPU (MIG) for secure workload isolation
  • Used for high-performance model serving

Learn more: NVIDIA Confidential Computing

privatemode.ai​

  • Confidential LLM inference service
  • No data retention or logging
  • Verified TEE deployment with attestation

Learn more: privatemode.ai Security

Further Reading​

For detailed implementation specifics, see the NVIDIA TEE and privatemode.ai sections.