Claude can now enter all your passwords for you, if you give him permission

1Password Introduces Partnership with Claude, Enabling Anthropic AI to Authenticate on Users’ Behalf Through Zero-Exposure Architecture
Users approve each biometric login.
New Agentic mode in browser extension crashes interface when AI agents take control

1Password, the leading password management company, has launched a new tool that allows artificial intelligence assistant Claude to authenticate on behalf of its user and thus complete tasks that were previously impossible without significant security trade-offs.

1Password for Claude is based on a “zero exposure architecture”, so in practice this means that Claude can basically ask 1Password to complete the login process, but he will never see the credentials and they will never be loaded into his memory.

In turn, 1Password will notify the user and request biometric approval before continuing. Once granted, it will automatically populate the credentials and check whether they were exposed on the page or not. If the send fails, the completed values will be cleared and reported.

Agent mode

“We need a new security model designed specifically for agents, not just humans,” said Nancy Wang, CTO at 1Password. “The answer is not to give your secrets to agents. It is to allow a user to give an agent permission to use a credential without allowing the agent to see it. Claude knows that he used his login; he does not need the password or the one-time code in his context. That distinction is where trust in agents begins and the foundation we are building with Anthropic.”

To further strengthen its security posture, 1Password also announced Agentic Mode, a new feature in the browser extension that gives users visibility and control over browser-based AI agents. When a supported AI agent takes control, the 1Password extension automatically locks and hides the interface. The agent can only use logins and OTPs that are explicitly approved for the current task.

Even if the integration is not configured, and even if 1Password is not needed for the current agent task, Agentic Mode works, the company emphasized. Other agents besides Claude also receive support.

An important debate is currently taking place about how many permissions AI agents should receive and under what rules. We’ve already seen horror stories of AI agents deleting all of people’s email inboxes or ruining days of hard work. It remains to be seen if this recovers or if most people remain skeptical about giving AI access to certain services.