The whole excuse OpenAI gave for going proprietary was specifically that they could spy on you and use that to reactively adjust GPT's alignment training. For

The whole excuse OpenAI gave for going proprietary was specifically that they could spy on you and use that to reactively adjust GPT's alignment training. For what it's worth, the US military is also evaluating LLMs in the same way the Chinese and Russian hackers are. I imagine - or at least hope - they aren't using OpenAI's hosted ChatGPT and are getting copies of models they can run on SIPRNET or JWICS.