Is it safe to give an AI job search agent my inbox?

It's as safe as the limits on what the agent can do, and Shep's are strict: it reads your inbox to find job email, and nothing it reads can send, delete or apply on its own.

Here are the limits, and how each one works:

• An email is data, not instructions. Anyone can email you, so a message can't tell Shep what to do. The model that sorts your mail is told to ignore any instructions inside it, and every card built from an email waits for your approval, in every mode.
• Nothing sends by itself. An email leaves only when you press Send, or approve that one email's card. Shep never deletes your mail.
• Only a little leaves your inbox. To sort job mail from the rest, each new message's sender, subject and preview text go to an AI model from OpenAI or Anthropic. Mail that isn't about a job is skipped, and Shep keeps only its message id, for 90 days.
• Your mailbox is yours alone. The connection belongs to your account. One mailbox can't be linked to two accounts, and the sign-in check makes sure the browser that finishes connecting is the one that started.
• Leaving takes one step. Disconnect in Settings and the access is revoked at Nylas, the email service the connection runs through. Delete everything removes your data and revokes the inbox first. If a trial ends, the inbox disconnects and what it filed stays.

Before you connect any AI agent to your inbox, ask four questions. Can it send without you? Can an email tell it what to do? What does it keep? How do you cut it off?

Trust sets out what Shep reads, keeps and never does.