Cold Start Attestation Latency in Serverless Confidential Inference
Six sequential layers, not one bottleneck, determine confidential inference cold start cost.
Kian Vance
Section
Confidential AI Inference at Scale
6 stories in Confidential AI Inference at Scale.
Latency and Throughput Overhead of TEE-Protected Inference
SGX's memory ceiling makes confidential VMs the safer choice for large models.
Lorena Sundqvist
Multi-Party Inference with Confidential Computing
TEEs offer the only production-speed path to private multi-party inference.
Kian Vance
Model Weight Protection Inside Confidential Inference Servers
Hardware enclaves protect model weights where policy and access controls cannot.
Lorena Sundqvist
Encrypting Prompts and Responses in Transit to a TEE
TEEs encrypt prompts inside enclaves, closing the gap TLS leaves behind.
Kian Vance
Threat Model for a Confidential LLM Inference Endpoint
Five trust boundaries structure risk in confidential LLM inference deployments.
Kian Vance