François Fleuret proposes 'AI Safety Levels' air-gapped facilities modeled on bio safety levels

francoisfleuret · x · 2026-09-20

Researcher François Fleuret floats an idea nobody is discussing: "AI Safety Levels" facilities mimicking bio safety levels, with real air gaps inside Faraday cages plus standard high-security controls.

Levels could be tied to the number of parameters or FLOPs of the models inside the facility — a concrete proposal to physically standardize frontier AI security.

Original post →

More from Infra

Infra channel →