A New Way for Claude to Access Your Online Accounts
1Password has introduced a way for Claude, Anthropic's AI agent, to log into your online accounts without your actual password ever becoming part of the exchange. The new 1Password for Claude integration lets the assistant fill in saved login details during a task while keeping those credentials completely hidden from Claude and from Anthropic's systems.
The feature is live now for Mac users. It activates the moment Claude lands on a sign-in page while carrying out a task on your behalf.
How the Login Request and Approval Process Works
Claude Asks, You Decide
When Claude hits a page that requires a login, it doesn't reach into your saved credentials on its own. It sends a request to 1Password asking to use a specific saved login, and you get the final say — approving or denying the request before anything else happens.
Credentials Travel Through a Separate Channel
Once you approve, 1Password doesn't hand your password to Claude. It submits the login information through its own encrypted channel, kept apart from whatever Claude is doing in the browser. Your password and any one-time codes never show up inside Claude's context, and they never reach Anthropic's servers either.
What Claude Actually Sees During Sign-In
Claude's part in this process is deliberately narrow. It identifies which site needs a login and puts in the access request, but what it gets back afterward is minimal — just confirmation of whether the attempt worked and which saved item was used. The actual decrypting and typing of the password happens inside the 1Password browser extension, and Claude pauses reading the page while that happens. That timing removes any window where the agent could inspect the credentials mid-entry.
Agentic Mode Adds Another Layer of Control
A separate setting called Agentic Mode switches on automatically whenever Claude takes control of a browser tab. It strips away 1Password's usual autofill interface and restricts the agent to only the vault items approved for that specific session — nothing else in the vault is accessible.
There's a built-in fail-safe too: if a login attempt doesn't succeed, the browser extension clears the credentials before handing control back to Claude. That prevents any situation where a half-completed login form is left sitting on screen for an AI agent to examine.
Where the Protection Stops — and Why That Matters
It's worth being precise about what this system actually covers. The safeguards in 1Password for Claude apply to how credentials are approved and delivered — not to everything that happens once you're signed in. After a successful login, Claude is operating inside that account the same way any authenticated user would. From there, what it can see or do depends on Claude's own controls and on whatever actions that particular website permits.
A secure sign-in process solves the problem of keeping your password away from an AI model. It doesn't, on its own, answer the separate question of how much you're willing to trust that agent once it's already inside your account.
System Requirements and Who Can Use It
1Password for Claude is available now to individual, family, and business subscribers — but only on Mac. Using it requires:
- Version 8.12.28 or later of the 1Password desktop app
- The matching version of the 1Password browser extension
- Claude's desktop and browser extensions installed
Business accounts have one extra step: administrators must turn on agentic autofill before employees can connect and use the feature.
Supported Login Types
This first release works with logins saved as usernames, passwords, and one-time codes. Passkeys aren't supported yet, and neither are social sign-in options such as Sign in with Google.
A Sensible Way to Start
Given those current limits, it's worth testing the integration on a low-risk account first, with a narrowly defined task. Hold off on connecting anything that would be expensive, public, or difficult to undo if something went wrong until you've seen how the system behaves in practice.