www.it-connect.fr
1Password Lets Claude Log In to Your Accounts Without Seeing Your Passwords
1Password and Anthropic are taking action: since July 16, 2026, the 1Password for Claude feature has allowed Anthropic’s AI assistant to sign in to your online accounts on your behalf, without ever seeing your passwords or one-time codes. An interesting answer for AI agents that can click, buy, and fill out forms, so they can fully act inside your browser.
Sign in without ever exposing the password
AI agents are able to act on our behalf in browsers and applications: Claude (and it is not the only one) can compare offers, add an item to the cart, and even go as far as completing a purchase. Except that at some point, the AI inevitably runs into a login page. Until now, you had two options, and neither was satisfying: give your password to the agent, or take back control and do the task yourself. A point of friction that 1Password wants to remove thanks to its integration with Claude.
1Password for Claude relies on a so-called zero-exposure architecture, meaning your precious credentials are never exposed. Claude can carry out tasks that require a login and a one-time code, but the credentials are never actually transmitted to the model, nor stored in its memory. 1Password remains the sole source of truth for the secret, and access is granted only at the moment the task is executed.
When Claude needs to sign in, 1Password shows the user which credential is being requested, and why. After you approve it via biometric authentication (Touch ID on Mac), 1Password injects the credential into the web page. Claude sees neither the vault item, nor the password, nor the one-time code. The access is isolated so that it is limited to the current task and ends once it is completed: Claude does not have persistent access to benefit from this autofill.
"Once autofill is complete, 1Password checks that the confidential information has not been exposed on the page. If submission fails, it clears the entered values before returning control," we can read. For its part, Claude knows that the sign-in has been established, without having had access to the sensitive information. This also means that you are putting all your eggs in one basket: passwords and TOTP code generation.
This mechanism works on sites where Claude can already act through Google Chrome. This reminds me of a recent security issue: the ClaudeBleed flaw discovered in the Claude in Chrome extension.
Agentic Mode: lock down the vault when AI takes over
What happens when an AI agent has access to the browser and takes control of a machine on which 1Password is installed? It could try to interact with the extension itself. This is where a feature called Agentic Mode, directly integrated into the 1Password extension, comes in.
The goal: make the AI agent blind, so that it cannot interact with the vault. "When a compatible AI agent takes over, the 1Password extension automatically locks. The interface is hidden, and the agent can use only the credentials and one-time codes explicitly approved for the current task. The rest of the vault remains inaccessible.", 1Password explains.
1Password says that Claude is not the only supported tool, without giving any other examples. In any case, 1Password is clearly determined to position itself at the access layer between AI agents and user credentials. This is an interesting and very real issue, so it is reassuring to see concrete proposals on the subject.
Well, if you want to take advantage of it, you must meet the following requirements:
- Available on Mac now (1Password does not mention other operating systems),
- Available with personal, family, and business plans.
- Have the 1Password desktop app, the 1Password browser extension, the Claude desktop app, and the Claude in Chrome extension.
Find more information in the 1Password documentation.
