1Password has introduced a new way for Anthropic's Claude to log in to websites without exposing the underlying secret to the AI model itself.
The product, called 1Password for Claude, is built around what the company describes as a zero-exposure architecture. In practice, that means Claude can use a person's credentials to complete a task, but it cannot read the password, inspect the vault item, or access the one-time code. 1Password remains the only holder of that information, and access is granted only for the duration of the approved task.
Credentials are injected only at the moment of use
According to 1Password's blog post, when Claude needs to sign in, 1Password first tells the user which credential will be used and why. Only after the user approves the action with biometric authentication such as Touch ID does 1Password fill the username and password directly into the webpage.
Claude does not get to see the password itself, the vault item, or the one-time verification code during that process. If the submission fails after the form is filled, 1Password clears the entered content before handing control back to the agent.
1Password chief technology officer Nancy Wang said, “We need a security model built for agents, not just humans... The answer is not to hand your secrets to the agent, but to let users authorize an agent to use a credential without letting the agent see it. Claude knows it used your login information, but it does not need to put your password or one-time code into its context.”
Mac-only at launch, with several software requirements
For now, 1Password for Claude is available only on Mac. It requires the 1Password desktop app, the 1Password browser extension version 8.12.28 or later, Claude Desktop, and Claude in Chrome. The feature is available across personal, family, and business plans.
Agentic Mode is meant to stop browser-level overreach
The zero-exposure model addresses one question: whether an agent should be given the password at all. But it leaves another one open. If the agent controls the browser, could it try to interfere with the 1Password extension itself?
To deal with that risk, 1Password has added Agentic Mode to its browser extension. Once a compatible AI agent takes control of the browser, the extension automatically hides and locks its interface. In that state, the agent can use only the login credentials and one-time codes that were explicitly approved for the task at hand. Everything else in the vault stays out of reach.
The safeguard does not require extra setup. It turns on automatically even if the current task does not directly use 1Password, and it is not limited to Claude alone. Other compatible agents are covered as well.
Initial support is narrow
The first release supports only login-type items, specifically usernames, passwords, and one-time verification codes. Passkeys, social logins, payment cards, and identity information are not yet available. The company said those areas will be added after launch.
Survey data points to a trust gap around payments
1Password also pointed to survey data from PYMNTS Intelligence. The survey found that 56% of online shoppers are willing to let AI agents help compare prices, while fewer than 40% are willing to let agents handle payment credentials. That gap helps explain the trust problem this design is trying to address: users may accept agents doing tasks on their behalf, but remain more cautious when money or identity data is involved.