Cryptographic identity for AI compute infrastructure and agentic AI systems
Explainable NVIDIA-focused GPU workload, attestation, attack-path, and AI-agent security lab
Cryptographically isolated C++ GPU execution runtime for NVIDIA Confidential Computing (Hopper/Blackwell). Features SPDM 1.2 remote attestation for driver/firmware verification and hardware-accelerated AES-256-GCM VRAM tensor decryption over PCIe buses.
Verified Inference Path for NVIDIA Triton. Protects GPU compute resources from unauthorized autonomous inference requests via DCC.
Complete security toolkit for enterprise NVIDIA GPU infrastructure. Includes NIST 800-53 controls, Zero Trust architecture, threat models, incident response playbooks, forensic scripts, and monitoring configurations for H100/A100/L40S and other datacenter GPUs.
Practical guardrails against silent GPU-side model corruption
Tenant-side, canary-only GPU assurance framework: measure memory sanitisation, hardware consistency and isolation on rented cloud GPUs, with signed evidence. Pre-alpha.
To associate your repository with the gpu-security topic, visit your repo's landing page and select "manage topics."