AI Agents Will Log In Without Seeing the Password: 1Password's New 'for Agents' Service
Getting an AI agent to shop online, book tickets or work on various websites usually requires logging into accounts with a password, along with personal information. In that process, the AI agent retains access to that information or password, which can create a security risk later. To solve this problem, password management company 1Password has launched a new service called '1Password for Agents'.
Through the new platform, AI agents will be able to enter various accounts and complete tasks according to the user's instructions, but will not be able to directly see or collect the stored passwords. In other words, the work gets done, but the secret information never passes into the agent's hands.
1Password is essentially a password manager, where users can securely store passwords and other confidential information for various websites and online services. Until now, to get an AI agent to work on a website, the user either had to log in to the account themselves, or type the password into the agent's chat.
In the second method, a copy of the password risks remaining with the AI system, a major security concern. To avoid that risk, the new 1Password platform will let AI agents enter accounts with the user's permission while keeping the password secret.
To log in to a website through '1Password for Agents', the AI agent will first request permission to use the relevant account's information. The user can review the request on their own device and see which account's information is needed and for what reason.
If permission is granted, 1Password will enter the password directly into the login form of the relevant website. As a result, the password never needs to be sent to the AI agent. Once the login is complete, the password will be kept out of the agent's reach, so it cannot read or reuse it.
Permission must be taken separately for each session. The permission is limited to the approved information of a specific account and does not remain valid in the next session. Passwords and one-time multi-factor authentication (MFA) codes are supplied through 1Password's controlled secure channel, so this information is not directly exposed to the AI model.
The company said 1Password can currently be used with Meta's Muse, Claude Duet, ChatGPT Work and Dots, Grok Bot and Town. For these AI agents, users will be able to grant permission to use specific account information as needed.
1Password claims that in this process, passwords, usernames and MFA codes stay out of the AI agent's reach. As a result, the risk of such confidential information leaking through AI models is reduced.
There will be no additional fee to use '1Password for Agents'. Customers with personal and family accounts on 1Password.com will be able to use the facility.
As the use of AI agents grows, questions are growing about the risk of handing them the keys to accounts. This model of logging in without revealing the password adds a new dimension to that debate. The technology sector will be watching to see whether other password management companies follow the same path in the coming days.
