The new feature allows approved AI agents to access accounts through 1Password while keeping the underlying credentials hidden from the software.
1Password has introduced a way for AI agents to sign in to user accounts without ever seeing the underlying password, in a move that reflects growing concern over how much access autonomous software should be given to personal services.
The feature, called 1Password for Agents, is designed for scenarios in which a model needs to complete a task on a website but should not be handed the user’s login details. According to 1Password, an agent that reaches a sign-in page can present a unique digital key to identify itself to the password manager. The user is then notified, told which credential is being requested and why, and can approve or reject the request.
If permission is granted, 1Password fills in the login form directly, rather than passing the password to the agent. In practice, that means the software can enter the account and continue the task, but it does so without learning the secret itself. That distinction matters, because once a password is exposed to an AI system, it may be copied, cached or otherwise placed outside the user’s control.
The company’s pitch is as much about limiting exposure as it is about convenience. Manually logging in every time an agent needs access would be cumbersome, particularly for people using these tools for routine work such as travel booking, account management or other repetitive tasks. Yet the security trade-off remains obvious: even if the agent cannot view the password, it can still act inside the account once granted access, so users must still judge the risk of each request carefully.
According to reporting by PCWorld and 9to5Google, the service is intended to work with a range of agent systems, including offerings associated with Anthropic, OpenAI, Meta, xAI and Town. 1Password says the feature is available on its Individual and Family plans, although access may depend on region, device and sign-in method. The company has also positioned the launch as part of a broader credential-broker approach, in which access is checked at runtime and limited to the specific approval given for a task.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
