Skip to content

Staying safe with AI

How Plavn guards against emails that try to instruct the AI, why links are removed from drafts, and what to check before you send.

You stay in control. Plavn’s AI only writes drafts and suggestions. It cannot send mail, change your settings or delete anything, so nothing leaves your account unless you press Send.

On top of that, Plavn has safeguards that keep the AI from being misled by what is written inside an email. They are described below, and your own look at a draft before you send it is the last check.

  • Every reply, task description, rewrite or suggestion is a draft or a suggestion. Nothing is sent until you press Send.
  • The AI cannot change your settings, delete mail or open links. It only returns text. Plavn’s servers do not visit links or load files that they find in your email.
  • The morning brief can only point to tasks and mails that really exist. The AI chooses among numbered items and never writes a title or a link itself, so a hostile subject line cannot become a link.

Emails, notes, task titles, commit messages and code are handed to the AI as data in a fixed format, with an instruction that whatever they say must not be obeyed. This makes it much harder for a message to take over, but it is not a guarantee, which is why the next two safeguards exist.

Section titled “Links and addresses are removed from drafts”

An injected instruction often asks the AI to include a link or an account to pay. So in replies, rewrites, follow-up drafts and writing help, Plavn removes every link, website name and email address the AI wrote, unless it also appears in text that is yours: your own mail and notes, the draft you are editing, your instruction notes, or the sender and recipient addresses. A link that only appears in the incoming email is not trusted.

You see what was removed: [link removed] or [address removed] appears in the draft. If you want the link, type it yourself after checking where it goes.

If an email contains text that tries to give instructions to an AI, the prepared reply is shown with a red notice: This email contains text that tries to instruct an AI. Read the reply with care before you send it. The email is still handled as data, and the notice is there to help you decide. It is a helpful signal, not a promise to catch everything, so keep reading replies before you send them.

  • The recipient and subject, especially on a forwarded or unexpected mail.
  • Any amount, date, address or promise in the reply. The AI is told not to invent facts, but it only knows the email and what Plavn found in your notes.
  • Anything the reply asks the recipient to do (click, pay, reply with something).
  • The red notice above, when it appears.

If something looks wrong, discard the draft and write the reply yourself.

What Plavn sends · Your controls · Review and send a prepared reply