Meta's new Muse AI got full access to one writer's personal Gmail account, and within a day it was drafting and sending email under the writer's direction.
The writer connected Muse, an agentic assistant that runs continuously rather than waiting for a prompt, to a personal Gmail inbox and asked it to flag anything urgent. Muse surfaced three messages worth a look, including a pitch from a PR contact about an embargoed product. When asked to draft a reply, Muse's first attempt overpromised, offering a call the writer never wanted. After a couple of rounds of edits, Muse sent a corrected version straight from Gmail, cc'ing the writer's work address as instructed.
That persistence is the real story. Assistants like ChatGPT, Gemini, and Claude are built to wait for a prompt, but Muse watches the inbox on its own and flags problems before you ask, which is either the most useful thing to happen to email triage in years or a genuine leap of faith in a company with an uneven privacy record. Meta says credentials sit in an isolated "Secure Credential Store," but that promise is doing a lot of the reassuring here.
An AI that reads your email and asks before acting is progress. An AI that reads your email and never stops reading it is a different conversation.