1Password lets AI agents use your passwords without seeing them
1Password has launched a feature that lets AI agents sign into online accounts without revealing the passwords to the agents. It aims to make agent-led tasks that require logins easier while keeping account access under the user’s control.
Each agent uses a unique digital key to identify itself, then 1Password asks the user to approve access on their own device and states which account is needed and why. If approved, 1Password fills in the login details; the agent can proceed without seeing the password. The system is designed to work with agents including Meta Muse, ChatGPT Work, Dots, Claude Duet, Grok Bot and Town, with more integrations planned.
In short
- AI agents can sign in without seeing passwords.
- Users approve each request on their own device.
- 1Password plans support for more agents.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
This could make useful AI tasks easier while preserving an important boundary: agents can use an account only after the user approves a specific request, and they do not receive the password itself. Unique agent keys and an explanation of which account is needed and why may give users more control than sharing credentials directly, while allowing access to be limited or withheld.
The case against
Even without seeing passwords, agents that can sign in may still take actions or access sensitive information on a user’s behalf, so approval prompts may not remove the risks of mistakes, misuse or compromised systems. The protection also depends on how reliably agent identities and requests are verified, and on users understanding what access entails; wider integration could make those safeguards harder to assess.