AI agents become a lot more useful when they can actually do things for you. Planning a vacation, preparing an expense report, or handling a recurring bill can require access to an account, though. And handing an AI your password is hardly an appealing way to get started.
1Password is addressing that problem with 1Password for Agents, a platform that lets personal AI assistants sign into accounts without seeing or storing the credentials. The company is working with Meta, Anthropic, OpenAI, SpaceXSI, and others to bring the capability to their agents.
The idea is straightforward. When an agent encounters a sign-in form, it identifies itself to 1Password using a unique digital key. You then see which credential it needs and why it is requesting access. If you approve on your own device, 1Password fills the credential directly into the form, allowing the assistant to continue its task.
That sounds like a more sensible approach than copying a password into a chat window. It also gives users a chance to review a login request before an assistant gets into an account.
“Agents are starting to do real work on our behalf. The old model where you hand over a credential and hope for the best doesn’t work,” said Aashish Tripathi, chief product officer at 1Password. “An agent should be able to do the job it’s been asked to do without getting permanent access to everything behind the account, and 1Password provides the trust layer that makes this possible across personal agent experiences.”
Among the named integrations are Meta’s Muse, Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots, SpaceXSI’s Grok Bot, and Town’s Townie assistants. These tools serve somewhat different purposes, but they run into the same practical obstacle. Much of what people need help with sits behind a login screen.
Muse can request account access while working through a user’s plans, for instance, while ChatGPT Work and dots can use approved sign-ins to complete tasks on websites. Town’s assistants work with email, calendars, and business tools, requesting the logins they need along the way.
For people who already use 1Password, there is an obvious convenience here. Their credentials are already saved, so they would not need to create a separate collection of passwords for an AI assistant or manually feed them into each service.
However, the partner list should not be mistaken for a promise that every integration is ready today. Anthropic says 1Password will “soon” complete sign-ins for Claude Duet, and 1Password has not provided a full availability schedule or pricing details in its announcement.
Other partners working on the platform include Browserbase, Browser Use, Dana, Instinct, KERNEL, Perplexity, Ramp, Tomo, and Wajo. 1Password says additional integrations will follow, potentially giving users more choice about which assistant they connect to their vault.
I like what 1Password is trying to do here. If AI companies expect us to let agents handle more of our daily lives, they need a reasonable way to access the services we use. Making that process familiar to password manager users could help people feel more comfortable trying it.
Still, there is a larger question once the login succeeds. What can the agent actually do inside the account? Keeping a password hidden does not prevent an assistant from making a mistake with information or controls it can access after signing in.
1Password explains how users approve a login, but it has shared less about how access ends or which actions require additional permission. Those are things I would want to understand before letting an agent loose in a sensitive account.
1Password for Agents could make delegating online tasks more practical, particularly for existing customers. Just remember that approving an assistant’s login and trusting its judgment are two separate decisions.
Support independent tech journalism
NERDS.xyz is independently owned and operated. If you enjoy my coverage of Linux, AI, hardware, cybersecurity, and tech culture, consider supporting the site on Ko-fi.
Support NERDS.xyz
