Your LLM Stack Is a 2008 Database With Better Marketing — Lovina Dmello, NVIDIA
Most production ML security breaches stem from basic infrastructure mistakes, not exotic AI attacks
“almost everything that is breaking in the production ML security isn't some exotic AI attack. It's the same boring infrastructure mistakes that we supposedly fixed years ago.”
NVIDIA deep learning infrastructure engineer Lavina D'Mello argues that the AI industry is repeating classic pre-2008 infrastructure security failures — exposed APIs, over-privileged accounts, public model weights — rather than facing novel AI-specific threats. She cites a 2023 incident where thousands of Ray distributed ML clusters were left open on the internet due to authentication being off by default, resulting in over $1B in exposure. The core insight is that ML systems broke the deterministic security assumptions that classic tooling was built around, leaving teams without adequate mental models to secure probabilistic, multi-tenant AI infrastructure.