1Password has launched a new feature called "1Password for Claude." It lets Anthropic's Claude chatbot log into websites on your behalf, but Claude never actually sees your passwords or your second-factor login codes. The company announced the feature on July 16, 2026, in a blog post by Mitchell Cohen and Horia Culea titled "1Password for Claude: Give Claude access without giving up your credentials" (The Verge).

The feature uses what 1Password calls a "zero-exposure security framework." Think of it like a restaurant where the kitchen staff prepares your meal through a pass-through window: the waiter can deliver your food, but they never go behind the counter where the ingredients are stored. Claude can use your saved logins to sign you in, but your actual passwords and one-time codes stay hidden from the AI at every step (The Verge).

You approve each login individually. Every time Claude needs a password, you get a prompt to approve or deny the request using your fingerprint or face scan. When Claude takes control of the browser, 1Password locks everything down except the specific login you approved for that task. After filling in your credentials, 1Password scans the page to make sure no password information is left sitting on the screen before handing control back to Claude (The Verge).

At launch, the feature only covers website logins. Support for payment cards and saved identity details will come later. It is available on Mac for 1Password business, family, and individual plans. You need the 1Password desktop app, the 1Password browser extension, the Claude desktop app, and the Claude browser extension all installed (The Verge).

This builds on a partnership between 1Password and Anthropic that started earlier in 2026. In March, 1Password launched a broader initiative called "Unified Access for AI Agent Security," which included integration with Anthropic's Claude browser extension, Cowork, and Claude Code (1Password press release, March 17, 2026). For developers, 1Password also offers a Claude Code plugin that lets the Claude command-line tool sign in using an Anthropic API key stored in 1Password, protected by fingerprint or face scan (1Password developer docs).

The reason this matters is that AI agents controlling web browsers face a real security risk. If an AI can see your password, that password could be stolen if the AI gets tricked by hidden instructions on a webpage, or if the AI behaves in an unexpected way. The old approach of just giving an AI your password and hoping it uses it correctly is what 1Password is trying to replace.

Here is how the zero-exposure approach works. Claude says it needs to log into a site. 1Password shows you a fingerprint or face scan prompt. Once you approve, 1Password types your password directly into the website's login box through a channel the AI cannot look at. Claude ends up signed in, but never held the password. The page scan afterward adds a second check, making sure no leftover password data is sitting in the webpage's code before Claude gets control back.

This is a different approach from simply giving an AI agent a key or a login token to use. The AI never holds the secret at any point. So if someone manages to trick the AI into doing something wrong, the damage is limited to what the AI can do with the one login you approved, not to the password itself. And because each login requires its own approval, a compromised AI cannot quietly collect your passwords over time.

There are some limitations at launch. It only works on Mac. You need to install four separate pieces of software to use it. And it only handles website logins for now, not payments or identity information.

The Claude Code plugin for developers is the more established part of this partnership. It solves a simpler problem: letting developers use the Claude command-line tool without storing their API key in a plain text file. That plugin uses the key stored in 1Password, protected by a fingerprint or face scan (1Password developer docs). The new browser feature takes that same idea and applies it to the much messier world of an AI navigating real websites.

The path from the March 2026 launch to this browser feature suggests 1Password wants to be the go-between for passwords whenever AI agents are involved. The open question is whether the per-task approval prompt will feel manageable in everyday use. If an AI agent needs to log into several services to complete one task, you could get several approval prompts back to back, and people getting tired of repeated security prompts is a well-known problem in technology design. How 1Password balances thorough security with everyday convenience will determine whether this approach catches on widely or stays a tool for people who are especially security-conscious.