AI agents can be asked to carry out tasks that involve signing into online accounts. 1Password’s new feature, 1Password for Agents, is designed to let them do that without handing the agents the passwords themselves. Each sign-in request still requires the user’s approval.
How an agent gets signed in
When an agent reaches a sign-in form, it uses a unique digital key to verify its identity to 1Password. The app then notifies the user about which account credentials the agent needs and why it needs them.
The user can approve the request on their own device. If approved, 1Password fills in the required login information. The agent can continue its task without seeing or handling the password.
That separation is the central idea: the agent can use an account for the approved sign-in, while the user remains involved in deciding whether credentials should be used. The approval flow is described as similar to two-factor authentication, with approval happening on the user’s device.
Approval remains part of the process
1Password for Agents does not grant bots unrestricted access to every account connected to a user’s 1Password. Instead, a sign-in request prompts a decision about the particular credentials the agent says it needs. That gives the user a point to review the request before 1Password supplies the login information.
This could make tasks requiring account access less cumbersome: a user can authorize a sign-in without copying a password into an agent or giving the agent the credential directly. The system still depends on the user deciding whether to approve each request, and on the agent explaining which account it wants to use and why.
Keeping the password from the agent changes what it receives during sign-in, but the agent still needs to access the account to carry out its approved task. The feature’s described controls center on identity verification and user approval; the source does not describe additional limits on what an agent can do once it is signed in.
Support spans several AI agents
The access system is designed to work with a range of popular AI agents. The article lists Meta Muse, ChatGPT Work and Dots, Claude Duet, Grok Bot, and Town as agents it is designed to support.
1Password says more partners are working to add support. Those named are Browserbase, Browser Use, Dana, Instinct, Kernel, Perplexity, Ramp, Tomo, and Wajo, with “additional integrations to come.” The list signals that the feature is intended to work across multiple agent experiences rather than just one.
A new way to delegate account tasks
For users, the proposed arrangement makes sign-in a permission decision rather than a password-sharing step. They can ask an agent to do work involving an account, review its request for credentials, and approve or decline it from their own device.
1Password CEO Aashish Tripathi described the feature as a way for an agent to do its assigned job without permanent access to everything behind an account. That framing points to the broader challenge the feature addresses: agents may need access to services to complete tasks, while users need a way to control when credentials are used.
Whether this approach fits a particular task depends on the agent’s support for 1Password for Agents and on the user being comfortable approving the requested account access. The key distinction is that 1Password supplies the login information after approval, while the agent itself does not see the password.