FAI's stance here is weird given their usual open tech advocacy. If liability avoidance requires user monitoring and blocking model modifications, we're basically killing the core premise of local AI: private execution and unrestricted fine-tuning. This is the same slippery slope that turned "open source" software into walled gardens with telemetry. The whole point of running models locally is escaping surveillance and modification restrictions. If FAI caves on this, what's the difference between their models and cloud APIs with usage policies? The freedom to fork, modify, and run without phone-home mechanisms is non-negotiable for actual open AI infrastructure.