1Password lets AI agents log into your accounts without ever seeing your password

Password manager 1Password launched “1Password for Agents” on October 22, 2026, a feature that lets AI agents log into a user’s personal accounts without the agent ever receiving the actual password. The launch arrives as banks and other services grapple with a wave of AI agents attempting to access customer accounts, sometimes using credentials handed over directly by users.

How the approval flow works

When an AI agent needs to log into an account on a user’s behalf, it sends a unique digital key to 1Password to verify its identity. The user then receives an alert about the login attempt and can approve or deny it in real time. Once approved, 1Password itself fills in the credentials to complete the sign-in — the agent carries out the task without the password ever passing through its hands or getting stored anywhere it controls.

1Password’s pitch centers on a specific risk: handing a password directly to an AI agent means a copy of that credential can end up stored in places the user no longer controls, available for reuse long after the original task is done. By keeping 1Password as the intermediary, the company argues, users can delegate account access without losing track of where their credentials actually live.

Which agents are supported

At launch, 1Password supports sign-in delegation for several named agents and assistants, including Anthropic’s Claude Duet, OpenAI’s Dots and ChatGPT Work, xAI’s Grok Bot, Meta’s Muse, and agents built by the AI assistant startup Town. The company says it is working with additional AI companies to expand the list of supported agents over time.

Why the timing matters

The launch lands the same week major U.S. banks disclosed they are actively trying to distinguish authorized AI agent activity from unauthorized access in customer accounts — and largely struggling to do so. Bankers have described finding agent activity they couldn’t immediately attribute to a specific user request as, in one banker’s words, “horrifying.” A Chase spokesperson put the bank’s position bluntly: customers “should not give their usernames and passwords to anyone,” including an AI agent.

That’s precisely the gap 1Password is positioning this feature to close. Rather than asking users to choose between full credential-sharing and no agent access at all, the company is betting that a verified, revocable, per-login approval flow becomes the default way people grant AI agents access to real accounts — financial or otherwise — as agentic tools move from novelty to daily use. As reported by TechCrunch, the framing isn’t that agents are inherently malicious, but that consumers are increasingly delegating real tasks, including financial ones, to automated tools faster than the infrastructure to do so safely has caught up.