One of the gateways to incorporating AI into daily life is giving it access to emails. I know people use it to stay organized, create task lists, send emails on their behalf, etc. I am tempted, but I am reluctant to give agents (apart from the one my email provider already uses) access to my emails because:
- emails can contain private information,
- the other party has not given me permission to share their info with companies like OpenAI and Anthropic, and
- I want to limit access and exposure to my personal information (whenever possible).
I'm not perfect, but I make a concerted effort to minimize the amount of personal information I share with models. When conducting research or analysis, I scrub, depersonalize, and redact as much as I can, and stick to publicly available information and data if identity is necessary for context.
Unfortunately, not everyone shares the same level of caution, and I imagine my personal details sent over email are floating around on the interwebs and being used to train models without my explicit consent.