Canadian password-management company 1Password has launched a platform that lets AI agents sign in to accounts without receiving the passwords, aiming to make delegated tasks safer while leaving users responsible for deciding what the software should be allowed to do.
Why Do AI Agents Need Account Access?
An AI agent can use tools and websites to complete tasks, rather than simply answer questions. However, useful work often requires access to private accounts, whether to collect receipts, organise travel or prepare an expense report.
That creates a practical security problem. For example, pasting a password into a conversation may let an agent proceed, but it also puts sensitive information into a system where copies could remain beyond the task. Taking over every login manually avoids that shortcut but interrupts the automation.
The new 1Password for Agents therefore handles the sign-in after the user approves access to a particular account, allowing the agent to continue its task without receiving the password. As chief product officer Aashish Tripathi explains, “The old model where you hand over a credential and hope for the best doesn’t work.”
How Does The Sign-In Work?
When a supported agent reaches a login page, it requests access through 1Password and uses a digital key to verify its identity. The user sees which login is needed and why, then approves or rejects the request on their own device.
If approved, 1Password fills the credentials directly into the website instead of sending them to the agent. The agent can then continue its task without receiving the password itself. Connecting an agent does not give it access to every saved login in the vault.
For example, an assistant gathering travel receipts could request access to an airline account. The user could then approve that login while keeping unrelated accounts unavailable. This makes the decision more specific than handing over credentials for the assistant to retain and reuse.
Which Agents Are Supported?
1Password says it’s introducing the service gradually, with connections to Meta’s Muse and Grok Bot rolling out first. Support for Claude Duet, ChatGPT Work, dots and Town’s Townie is listed as coming soon.
The new personal integrations apparently come at no extra cost with Individual and Family accounts on 1Password.com. Accounts hosted on 1Password.eu or 1Password.ca aren’t currently supported, meaning UK users will need to check which service holds their account before trying to connect an agent.
Businesses will also need to confirm which features and administrative controls are available under their plan. Support for an agent doesn’t necessarily include every safeguard an organisation needs, and 1Password says, “The safeguards and controls vary by integration.”
What Protects The Password?
1Password’s documentation about its existing Claude integration provides an example of the kinds of safeguards it uses. For example, it describes checking the application’s identity, requiring human approval and encrypting communication between the desktop application and browser extension.
Before entering a password, the browser extension checks that the website matches the address saved with that login. The agent then temporarily stops reading the page while 1Password fills in the details and submits the form, allowing it to resume work after sign-in without receiving the password.
Does A Protected Login Make Every Action Safe?
Protecting the password doesn’t necessarily limit what an agent can do once it has signed in. Depending on the account’s permissions, it may be able to read information, change records or make purchases, even when the user has asked it to complete a much narrower task.
For example, an assistant collecting invoices may also have access to payment settings within the same account. Approving the login would let it reach those invoices, but would not automatically prevent it from changing billing details.
1Password’s FAQ acknowledges that what happens after sign-in depends on the agent and the website. Users therefore need to check what an account allows before granting access, rather than assume the agent will only be able to perform the requested task.
What Does This Mean For Your Business?
For businesses, a useful first trial would involve a clearly defined task and an account with access limited to what that task requires. Staff could then check the results and assess whether the sign-in requests explain clearly enough which account the agent needs and why.
If the agent can spend money, send messages or change important records, the systems it uses should require approval for those actions and enforce any relevant limits. This would give the business protection beyond simply telling the AI what it should and should not do.
There should also be a tested way to stop the agent and close any accounts it remains signed in to. Removing its connection to 1Password may prevent further sign-ins, but businesses should check whether existing website sessions need to be ended separately.
The launch of the 1Password system could make agents more useful by letting them access accounts without being given passwords. However, safer sign-in is only part of the solution. Businesses will still need to control what happens inside those accounts, with clear permissions and someone responsible for checking the work.
