1Password for Agents lets AI sign in without seeing passwords

Key takeaways

• 1Password launched 1Password for Agents on 22 October 2026. AI agents can sign in to a user's accounts, but the user has to approve each request, and 1Password fills in the credentials so the agent never sees them.
• Named launch partners include Meta's Muse, OpenAI's ChatGPT Work and Dots, Grok Bot, Town's agents and Anthropic's Claude Duet. Reports differ on which of these work today and which are "coming soon".
• The launch takes the Claude-only browser integration from July, which runs on Mac only, and opens it to other agent makers. Each grant covers one session and only the items the user approved.
• 1Password has not published pricing or rollout dates. Supported browsers and platforms for the new partners are also unclear.
• The launch comes days after JPMorgan Chase and Charles Schwab said they block outside agents that log in with customer credentials.

1Password for Agents, launched on Thursday, 22 October, lets personal AI agents sign in to websites only after the user approves each request. 1Password types the credentials into the login form, so the agent never sees them.

The Toronto-based password manager is betting that AI agents need a safer way to sign in than being handed a password. Today many agents get access because a user pastes a password into a chat or saves it in the agent's settings. 1Password argues that this leaves copies of the secret in places the user no longer controls, according to the company's statement quoted by TechCrunch.

How 1Password for Agents works

The flow below is drawn from TechCrunch, Engadget and The Next Web reports on the launch:

1. The agent identifies itself. When the agent hits a login page during a task, it presents its own digital key to 1Password. The key proves which agent is asking.
2. The user sees the request. 1Password shows which agent wants which login, and why.
3. The user decides on their own device. The user approves or rejects the request. Nothing is filled until they approve.
4. 1Password fills the form. After approval, 1Password enters the username, password and any one-time code directly into the site's login form through a channel it manages. The credentials stay in the user's vault, never in the model.

Engadget reports that access lasts for one session and covers only the approved items. It does not carry over to later sessions. Aashish Tripathi, 1Password's chief product officer, said agents should be able to finish a task "without getting permanent access to everything behind the account," The Next Web reported.

Which AI agents are supported

Every report names the same partners, but they disagree on what is live today. Engadget says 1Password "now works with" all of them. The Next Web says Muse is already requesting logins and that sign-ins for Claude Duet are coming soon. 1Password has not published a dated rollout schedule, so readers should treat the list as partners, not a guarantee that every agent works today.

Sources: TechCrunch, Engadget, The Next Web, 22 October 2026.

The Next Web also reports that browser-automation companies Browserbase and Browser Use, plus Perplexity, the corporate-card company Ramp, and five companies 1Password did not name have signed up. Nat Friedman, head of product at Meta Superintelligence Labs, said Meta was "so happy to have it integrated directly into Muse."

What is new compared with the July Claude integration

The idea is not new for 1Password. In October 2025 it released Secure Agentic Autofill in early access with Browserbase. On 16 July 2026 it launched 1Password for Claude, a narrower consumer version, which was covered on the 1Password blog and by MacRumors. That integration needs the 1Password desktop app and browser extension, the Claude desktop app and the Claude in Chrome extension. It runs only on Mac and requires a paid Claude plan. Attackers are using agents too: Microsoft says Storm-0940 ran the ARTEX AI pentesting tool against Microsoft 365 tenants.

The July release also introduced Agentic Mode in the 1Password browser extension. When a browser-based AI agent takes control, the extension locks down and hides its interface. The agent can then use only the logins and one-time codes approved for the current task. 1Password says this protection also applies to agents other than Claude, even when the Claude integration is not set up.

Sources: 1Password blog (16 Jul 2026), MacRumors (16 Jul 2026), The Next Web, Engadget (22 Oct 2026).

The main change is who can join. In July, one model maker reached 1Password through one browser extension on one operating system. The new platform has each agent prove its identity with its own key. Any company that adopts that scheme can then ask for credentials, so the agent's identity no longer depends on running inside a single Chrome extension.

What 1Password for Agents still cannot do

• It does not stop an agent's mistakes. Engadget notes that an agent that cannot reveal a password can still order the wrong product or book the wrong trip. Approving a login is not the same as approving every action taken after it.
• There is no published rollout or pricing. Neither the company nor the launch reports gave prices or partner-by-partner dates.
• Platform details are missing. The July integration was Mac-only. Reports on the new platform do not say which operating systems or browsers it supports.
• Sites can still refuse. 1Password controls how the password is entered, not whether a website accepts an automated login.

Why the timing matters: banks are shutting agents out

The launch lands in the same fortnight that large US financial firms drew a hard line on AI agents. JPMorgan Chase CEO Jamie Dimon said on the bank's 13 October earnings call that it will not let "just any outside agent" into the company, Fortune reported. A bank spokesperson said JPMorgan refuses third-party agents that use customers' banking credentials and supports customer-authorised data sharing through open-banking providers instead.

Charles Schwab CEO Rick Wurster told Bloomberg Television on 16 October that the firm blocks unauthorised agents, including Meta's Muse, that try to move client money. "We can't control it, the client can't control it, something could go wrong and who's liable?" he said, according to The Next Web. Schwab is offering read-only access through its own model context protocol (MCP) servers instead. TechCrunch, citing Axios, reported that a Chase spokesperson said customers "should not give their usernames and passwords to anyone." Bullquill's earlier report covers which US banks block AI agents and what they still allow.

The bank statements target agents that log in with a customer's credentials. 1Password's product changes who holds those credentials. It still ends with an automated agent signed in to an account. Neither JPMorgan nor Schwab has said publicly whether a 1Password-mediated login would be treated any differently. Banks so far favour channels they control, such as open-banking connectors and MCP servers.

What it means

For everyday users, the clearest benefit is that a password typed into a chat window can be leaked, logged or reused later, while a password that never leaves the vault cannot. The approval step also gives the user a record of which agent asked for which account, and why.

For the agent industry, the launch is part of a wider push to give agents their own verifiable identity rather than letting them pose as the user. Meta is building a personal agent protocol called "Poppy" with Sierra to cover how an agent identifies itself and gets customer permission, Fortune reported. 1Password's agent keys tackle the same problem from the password-manager side.

The weak point is acceptance. An approved, password-free login helps only if the website on the other end is willing to let an agent in. Banks and brokers have so far said no, which means agent logins are likely to grow first at lower-risk services such as shopping, travel and subscriptions. If you use an agent today, the safer pattern matches the advice in our guide to setting up passkeys: keep secrets in a manager you control and never paste them into an AI chat. For background on the agents involved, see our explainer on Anthropic's Claude Duet, how developers build an MCP server in Python, and our report on OpenAI agents and internal monitoring.

Sources

• TechCrunch: 1Password launches a way for AI agents to log in on your behalf without seeing your password (22 Oct 2026)
• Engadget: 1Password will let Muse, Dots and Grok Bot sign in to sites for you without getting your credentials (22 Oct 2026)
• The Next Web: 1Password lets AI agents sign in for users without seeing their passwords (22 Oct 2026)
• 1Password blog: 1Password for Claude (16 Jul 2026)
• MacRumors: 1Password adds Claude support (16 Jul 2026)
• Fortune: Jamie Dimon on AI agents, Meta's Muse and JPMorgan (15 Oct 2026)
• The Next Web: Schwab and JPMorgan are blocking AI agents from moving customer money