An AI agent that orders your supplies, books a table or files an expense report keeps running into the same wall: the login page. 1Password has built a way past it that does not involve giving the agent your password.
Until now, agents stopped at every login
When an agent reaches a sign-in page, it stops and asks for a credential. You had two choices. You could take over the session and type your username and password yourself, every time. Or you could paste the password into the prompt, which hands the agent a secret it can keep using after the task is done.
Passwords saved in a browser do not help, because an agent cannot use them. So an agent that needs ten different sites in a week stops ten times, and the work you wanted off your desk comes back to it ten times.
How the hand-off works
The agent has its own digital key, and it uses that key to prove to 1Password who it is. 1Password then sends you a notification that names the login the agent wants and the reason. You approve it on your own device, the way you approve a two-factor sign-in. Then 1Password fills in the form on the website, and the agent carries on with its task. The password, and any one-time code, never passes through the agent.
That has a practical effect. A password that never enters the prompt cannot be repeated in a chat, kept in a log, or sent to a stranger by an agent that has been tricked with a hidden instruction on a web page. And because the login stays in one vault, changing a password means updating it once, not in every tool that uses it.
A signed-in agent is the new thing to watch
We wrote earlier that an agent carries the keys of whoever set it up. This takes the password out of that bundle and leaves the access. Once the agent is signed in, it can act as you on that site.
Take a shop owner whose agent compares prices on three supplier sites and places the weekly order. If the owner approves the supplier login, the agent can place orders, and nobody can steal a password from it. If a page it reads tells it to order something else, the password is still safe but the order is not. The approval is what protects you now, so use it for what it is. Approve the supplier login for the supplier job. Say no when the reason line does not match what you asked for. Keep logins that can move money out of an agent's reach unless the task needs them.
Who can use it, and what came before
The feature is free for Individual and Family accounts on 1Password.com. Accounts on 1Password.eu and 1Password.ca are not supported. Meta's Muse and Grok Bot are first to get it, while Claude Duet and ChatGPT Work and dots are listed as coming soon.
The idea itself is a year old for developers. 1Password began testing Secure Agentic Autofill with the browser-automation service Browserbase in October 2025. Bitwarden published an open standard for agent credential access in March 2026. What is new is that someone with a family plan can use it with the assistants they already have.
When Apple Pay arrived, nobody asked shoppers to understand tokens. They only had to confirm the payment. With agents, the equivalent is the approval tap, and it is worth reading the reason line before you make it.