Right... I want less parties involved with secure data, not more. This should be an on-prem solution with no external network access and no direct calls to Op

Right... I want less parties involved with secure data, not more. This should be an on-prem solution with no external network access and no direct calls to OpenAI. A call is made to this service to obfuscate, then another call to OpenAI, all managed by a coordinating mechanism that is opensource / trusted. Better yet, maybe LLMs should be required to have weights released considering they are trained on the collective of human knowledge. Seems strange to use a significant sum of human knowledge that is publicly available then deny everyone access to the weights.