1Password for Agents, launched on Thursday 22 October, is a new way to let a personal AI agent sign in to websites for you without handing it your password. The agent asks, you approve on your own device, and 1Password types the credential straight into the login form. The model never receives it.
The launch arrives with a long list of partners. Anthropic's Claude Duet, OpenAI's ChatGPT Work and dots, Meta's Muse, SpaceXSI's Grok Bot and Town's Townie assistant are all named as supported agents. That matters because agents are already doing real tasks with real accounts: on the same day, Axios reported that US banks have started seeing agent activity inside customer accounts and often cannot tell whether the customer asked for it.
How 1Password for Agents works
1Password describes the product as a "trust layer" between you and your agent. Its press release sets out five steps:
- The agent reaches a sign-in page and proves its identity to 1Password with a unique digital key.
- You see which login the agent wants and why.
- You approve or decline on your own phone or computer.
- 1Password fills the username and password into the form directly.
- The agent carries on with the task. The password stays in your vault.
The design answers a habit that security teams dislike. "Handing over a password, secret, or credential is a common shortcut to an agent," 1Password wrote in its announcement blog. Pasting a password into a chat window means it can end up in logs, in the model provider's systems, or in the agent's memory. Here the agent gets a session, not a secret.
Which AI agents support it
- Claude Duet (Anthropic). Ami Vora, Anthropic's chief product officer, said "Claude Duet asks before it signs in anywhere," and that 1Password will soon complete sign-ins using passwords you already have saved. We explained Duet in our earlier guide, Claude Duet explained.
- ChatGPT Work and dots (OpenAI). Glen Coates, head of app platform at OpenAI, said the integration gives agents access to saved accounts without exposing credentials to the model.
- Muse (Meta). Nat Friedman, head of product at Meta Superintelligence Labs, described Muse as an assistant that gets safe access to services at the user's direction.
- Grok Bot (SpaceXSI), which can sign in without users pasting passwords, according to SpaceXSI's Tido Carriero.
- Townie (Town), where employees approve the logins a Townie needs.
1Password also lists Browserbase, Browser Use, Dana, Instinct, KERNEL, Perplexity, Ramp, Tomo and Wajo as partners, with more integrations "to come".
What it costs and who can use it
Neither the press release nor TechCrunch's report gives a price or a list of eligible plans. 1Password points users to an integrations page for availability details. If you already pay for 1Password, check that page before assuming your plan is included. The release also does not mention regional limits, so readers in Nigeria, Kenya, Ghana or South Africa should confirm in the app rather than rely on a launch announcement.
For scale, 1Password says its enterprise vault protects more than 1.6 billion credentials and secrets, and that more than 200,000 businesses use it.
What it will not do
The tool solves one problem: the agent never sees your password. It does not stop an agent from making a bad decision once it is logged in. If you approve a sign-in to your bank and the agent then misreads a task, the approval step is your last check. A few practical points follow from that:
- Read the request. The approval screen tells you which credential and why. Decline anything that does not match the task you gave.
- Keep money accounts separate. Banks such as Chase have told Axios that customers "should not give their usernames and passwords to anyone." Some banks, including Schwab and JPMorgan, already block outside agents from moving money.
- Use two-factor authentication. An agent session that hits a 2FA prompt still needs you, which is a useful brake.
- Review agent activity. Most agents now keep a task log. Check it after anything involving payments or personal data.
Why this matters for everyday AI users
Agents became useful this year because they can act, not only answer. Acting means logging in, and logging in has been the weak spot. A password manager sitting between the agent and the login form is a sensible fix, and the size of the partner list suggests the big labs agree.
Aashish Tripathi, 1Password's chief product officer, put it plainly: "The old model where you hand over a credential and hope for the best doesn't work." For anyone already letting an AI book, buy or file things on their behalf, that is a fair description of how most people have been doing it until now.
Would you let an AI agent log in to your accounts if it never saw the password? Tell us on our social channels.
