# multi-gpu

Distributed Training Attack Surface

Security vulnerabilities in multi-GPU, multi-node LLM training: gradient sharing attacks, parameter server compromise, insider threats, and infrastructure-level training exploits.