Does OPM/DOGE have a securely hosted model that is trained to do this already? Which model(s) and where/how they are hosted seems like a critical detail here.

Does OPM/DOGE have a securely hosted model that is trained to do this already? Which model(s) and where/how they are hosted seems like a critical detail here. Someone just needs to break into the email server or the LLM server to have a centralized view of all federal work. In isolation, assuming no one accidentally reveals sensitive information in an entirely new process without clear guidelines (lol), maybe the risk is low. In aggregate, information can become sensitive when it paints a broader picture. Without any doubt, Russia and China (among others) have set gaining access to this as a top priority.