"Would you let an AI read your email?" is the wrong question, in the same way "would you let a company hold your money?" is the wrong question. You already trust your mail to Google's or Microsoft's machines, which scan it for spam, malware, and search indexing. The real question is which additional party you're admitting, under what terms. Terms are checkable. Here are the seven checks.
The seven checks
| Check | What good looks like |
|---|---|
| 1. Training use | Your mail is never used to train models. Stated plainly in the privacy policy, not implied. |
| 2. Security audit | SOC 2 Type II or ISO 27001. An external auditor verified the controls; the vendor didn't just describe them. |
| 3. Access mechanism | Official OAuth via Google or Microsoft. No password sharing, ever, and access is revocable from your account settings in one click. |
| 4. Deletion | Disconnecting deletes your processed data, on a stated timeline. "Deleted" means deleted, in writing. |
| 5. Sending control | Draft-and-approve. The AI writes; you review and send. Nothing leaves your account without your click. |
| 6. Policy readability | You can find and understand the data handling section in five minutes. Impenetrability is a signal. |
| 7. Jurisdiction | GDPR compliance if you're in Europe, with a data processing agreement available for business use. |
How to verify without being a lawyer
Three sources, fifteen minutes: the vendor's security page (certifications are always displayed when they exist; absence is an answer), the privacy policy's data-use section (search it for "train"), and your own Google or Microsoft security settings after connecting, where the OAuth grant appears and where one click revokes it. If any of the seven can't be verified in those fifteen minutes, that's your answer.
The honest risk framing
A tool passing all seven checks handles your mail with more explicit controls than most human assistants ever operated under, and the value trade is the same one executives have always made: access in exchange for hours. Pidgy is built to pass this list (OAuth-only access, no training on your mail, draft-and-approve on every send, clean deletion on disconnect), and the same checks apply to any tool in the category. What machines can do with the access, once granted safely, is covered in what to delegate and what to keep.
