AI agents are more capable than ever, but even if you're comfortable with one completing a purchase on your behalf or booking a hotel, do you really want to give them a password? If not, you're not alone—68% of Americans wouldn't let AI act without their specific approval, according to research from YouGov, in part because it can be manipulated by prompt engineering attacks, and leave users legally liable for their behavior.
That's why password manager 1Password's new tool, 1Password for Claude, allows Anthropic's chatbot to complete browser tasks that require logins and one-time passcodes, without ever requiring you to enter your credentials into the AI model or store them in its memory.
For example, a small business owner could ask Claude to create a revenue summary or to flag any unusual activity using data from popular payment solutions like Stripe. Claude can then navigate the dashboard independently, and the business owner can approve Claude to use their Stripe login. 1Password handles credentials and one-time passcodes, so the task is completed without ever manually logging into Stripe or handing an LLM your sensitive login details. 1Password has included security features to mitigate the risks associated with agentic browsers, in which an agent could gain control of a browser with 1Password already installed.
Agentic Mode, a new feature in the 1Password browser extension, means the password manager shuts down when a compatible AI agent takes over, locking down 1Password. With Agentic Mode, the interface is hidden, and the agent can only use the logins and one-time passcodes explicitly approved for the current task.
1Password for Claude is now available on Mac across its business, family, and individual plans. For instructions on how to get the integration up and running, head to 1Password’s documentation page. You'll need 1Password for Mac (version 8.12.28 or later), 1Password browser extension (version 8.12.28 or later), the Claude desktop app, and the Claude browser extension.