1password Lets AI Agents Sign In Without Seeing Credentials
The new 1Password for Agents lets users approve individual agent sign-ins while keeping passwords and other login details outside the AI model.
1Password has launched 1Password for Agents, a system that lets AI assistants sign in to users’ accounts to carry out tasks without receiving their login credentials. Users approve each requested sign-in, while 1Password supplies the information needed to complete it, according to TechCrunch and Engadget.
Table of Contents
How an agent gets access
When an agent reaches a login screen during a task, it sends a special key to 1Password to verify its identity. The user then receives a notification about the request and can decide whether to approve it. If approved, 1Password fills in the required login information rather than handing it to the AI model, the reports say.
Engadget says approval applies to a particular session and specific items; access does not carry over to later sessions. It also reports that usernames, passwords and one-time multifactor authentication codes are supplied through a 1Password-managed channel, keeping them outside the model and preventing the agent from seeing the contents of the user’s vault.
Supported assistants and intended uses
Both outlets list Claude Duet, OpenAI’s Dots and ChatGPT Work, Meta’s Muse, Grok Bot and agents from Town among the supported assistants. TechCrunch says 1Password is working with other companies to expand the service. Engadget notes that 1Password introduced a Claude integration in July, before this broader launch.
1Password presents the system as a way to let personal assistants handle tasks such as restaurant bookings, gift purchases and vacation planning, according to Engadget. Such tasks may require signing in to a user’s account, but giving an agent a password directly could leave a copy of that credential somewhere the user cannot control, 1Password said in a company post cited by TechCrunch.
Limits and remaining risks
Keeping credentials hidden does not guarantee that an agent will carry out a task correctly. Engadget cautions that an assistant with approved account access could still order the wrong product or make an unwanted booking. The outlet also says 1Password had not disclosed pricing or detailed availability information.
The launch comes amid broader questions about how services recognize activity performed on a person’s behalf. TechCrunch cites an Axios report that U.S. banks have detected AI-agent activity in customer accounts, while many cannot determine whether that activity follows a customer’s instructions. The new 1Password system addresses how credentials are shared with agents; it does not, by itself, resolve what an agent might do after a user approves a sign-in.
Sources
This story was compiled by AI from the reports below. Read the originals for the full details.