1Password lets AI agents log in for you without ever seeing your password

Password manager 1Password released a feature called 1Password for Agents on October 22, allowing users to let an AI agent sign into their personal accounts on their behalf — without the agent ever holding the password. TechCrunch framed the launch against a wider problem: consumers are handing more account access to agents, and security vendors are trying to catch up.

The flow is a three-way conversation between the agent, the user and the password manager. When an agent tries to log in, it sends a unique digital key to 1Password to prove its identity. The user then gets an alert about the attempt and has to confirm it personally; only then does 1Password sign the agent in, filling in the credentials while the agent gets on with the task at hand.

Users can choose which agent they delegate to. The supported list includes Anthropic's Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta's Muse and agents from the AI assistant startup Town. 1Password says it is working with other companies to widen that list.

In a blog post, the company argued that handing a password, secret or credential straight to an agent is a common shortcut, and one that leaves sensitive information in places people can no longer control once a copy is saved for next time. 1Password for Agents is meant to offer a trusted way to delegate the task while keeping control of the passwords, secrets and credentials it requires.

The timing is not incidental. On the same day, Axios reported that U.S. banks have begun detecting agentic activity inside customer accounts. That activity is not necessarily rogue software — it is often consumers letting automated services act for them, including on financial matters — but most banks cannot tell whether an agent is acting on the user's instructions. Bankers described the situation to Axios as horrifying, and a Chase spokesperson said customers should not give their usernames and passwords to anyone.

That is why the login step has become the blocking point for agents. A year of agent demos has been about planning and tool use, but the moment an agent needs to click pay, submit or send, the obstacle is rarely model intelligence — it is identity and authorization, on websites built for humans with passwords, CAPTCHAs and risk engines. Keeping credentials inside the vault and requiring the user to approve each login is least-privilege design arriving in a consumer product.

What to watch: whether ordinary websites accept agent-initiated logins without flagging them as fraud, how wide 1Password's agent partnerships become, and whether regulated sectors such as banking will honour an agent action a customer has explicitly approved. The report did not detail which sites are supported or how the feature will be priced, and those details will decide whether this is a demo or a daily-use front door.