1Password lets AI agents log in without seeing passwords
Users can authorise an agent to use an account without showing the login details to the model.
Listen to this article
There is no audio version yet. Request one and an AI voice will read the article aloud.
Read by an AI voice.
Password manager 1Password has launched a platform that allows AI agents to log in on behalf of users without seeing or storing the password themselves. The user must approve every access request.
An AI agent can use the new service to make a login request while carrying out a task, such as a booking or administrative procedure. According to the company, 1Password verifies the agent’s identity and enters the details directly on the website.
The user can see which login is required and approve or refuse it. According to 1Password, the password remains in the service’s vault and is not passed to the AI model. The company is thereby seeking to limit a risk that arises when people enter passwords directly into a chatbot or agent.
The service initially works with a limited number of partners. Integrations with Muse and Grok Bot, among others, are being rolled out, according to 1Password; links with Claude Duet, ChatGPT Work and other agents have been announced as upcoming or in preparation. Availability varies by partner and account type.
The approach does not eliminate all the risks posed by agents. An agent can still carry out the wrong action, a user can approve a request incorrectly and a website can process the action differently from what was expected. Security also remains dependent on the software, the integration and the control of the device on which permission is granted.
The launch forms part of a broader shift in which AI systems not only provide answers but also operate websites and business services. For companies, this raises new questions about permissions, logging, liability and withdrawing access. 1Password says the new feature is available at no extra cost for certain individual and family accounts, but not all European account variants are supported immediately.
Also show the region and a close-up
Fact-check Approved · Nour Haddad — AI agent
This check was carried out by AI: every claim was re-tested against the sources. Even an approved article can contain errors — stay critical.
The launch, operation and announced integrations are based on the primary company source and an independent tech publication. The text presents security benefits as claims by 1Password and identifies remaining risks.
- confirmed 1Password launched a platform for AI agents on 22 October 2026. — This is stated in 1Password’s official announcement. source
- confirmed Agents can log in without showing the password to the model. — 1Password and MacRumors describe the product as working this way. source
- confirmed Security depends on software, integrations and user approval. — This is an inference from the described operation and is presented as a risk, not as the outcome of an independent test. source