Claude can now sign into your accounts without seeing your passwords, thanks to 1Password 🔐

Cliquez ici pour lire en français

Anthropic and 1Password have formalized a partnership that changes how an AI agent interacts with digital credentials. Called « 1Password for Claude, » the integration lets the assistant sign into websites to complete tasks, without the password, one-time code, or any other secret ever reaching the model, its memory, or Anthropic’s systems.

A « zero-exposure » architecture 🛡️

Here’s how it works: when Claude needs to authenticate on a site, 1Password shows the user exactly which credential is being requested and why. An explicit user approval then triggers the direct injection of the password or one-time code into the web page — Claude never sees it. During that operation, Claude actually stops reading and tracking the page until 1Password reports back; by the time the agent looks at the page again, the credentials have already been submitted and are gone from its field of view. That temporary blind spot, rather than any kind of encryption trick, is what keeps the secret out of the model’s context entirely.

Agentic Mode: a shield for every browser 🧩

At the same time, 1Password is rolling out a new feature called Agentic Mode, available to all its users. The moment a compatible AI agent takes control of the browser, the 1Password extension locks itself down and hides its interface. The agent can then only reach the credentials explicitly approved for the current task — everything else in the vault stays out of bounds. This protection kicks in even without setting up the Claude integration first, and 1Password says it plans to extend it to other AI agents as the ecosystem grows.

A new definition of digital trust 🤝

Nancy Wang, 1Password’s CTO, sums up the stakes: security needs a model built for agents, not just for humans. As she put it, the goal isn’t to hand an agent your secrets — it’s to let it use a credential without ever seeing it. Claude knows a login happened; it never needs the password itself in its context.

Currently available on Mac, the integration requires the Claude desktop app plus the Claude in Chrome extension, and is limited to paid plans (Pro, Max, Team, and Enterprise). It supports logins and one-time passcodes; credit cards and ID documents reportedly aren’t covered yet. It’s a telling sign of how far agentic AI has come — and of the tighter governance that comes with giving it real access.

Would you trust an AI agent to sign into your accounts on your behalf, even without it ever seeing your password? Let us know in the comments.