I’m only sending GPT message snippets, which Gmail’s API make available - probably using a language model (LoL). So there is very little PII in the API calls to GPT. That being said, I think OpenAI has a very large incentive to carefully destroy any query data sent through its APIs. The last thing they need is for some employee to quit and spill the beans about how Sam Altman stays up late at night laughing at all the API calls revealing the trivial problems of humans with an IQ of less than 140.