1Password launches a sign-in tool for AI agents

By VK

Published 26 Oct 2026, 12:22 UTC

Reporting from TechCrunch

1Password for Agents lets AI agents such as Muse and Grok Bot sign in once a user approves, filling logins 1Password says the agent never sees.

What it means for founders

Agent builders now have a login partner to weigh. 1Password invites agent makers to integrate, and Meta, Anthropic and OpenAI are already on its list. If your product signs users in to third-party sites, it may be worth comparing a credential broker like this with asking users to paste passwords, a practice 1Password argues can leave copies beyond their control.
Hiding the password may not limit the session. By 1Password's own account, an approved agent may still act inside the account. Teams running agents on user accounts may want their own limits on what happens after login, and records users can check.
Access is still not guaranteed. Companies keep tightening the rules for agents, from Delta, Yelp and Chase turning agents away to Apple's new consent step for Full Disk Access. A valid login may still meet a site that refuses automated visitors.
Watch the rollout. The signals are whether Muse and Grok Bot support arrives in the promised days, when the coming soon partners go live, and whether any bank says it will accept agent logins brokered this way.

The story

On Thursday, October 22, 1Password introduced 1Password for Agents, a service meant to get personal AI agents past login pages without users pasting a password into a chat. The company's announcement pitches it as a trust layer for agents, and TechCrunch covered it as a security story about the accounts people now let agents run.

How 1Password says the sign-in works

1Password describes three parties in each sign-in. The agent identifies itself to 1Password with a key of its own. The person, not the agent, then approves or declines on their own device, after seeing which login is wanted and why. If they say yes, 1Password enters the login itself, one-time code included, so the agent, in the company's telling, is told only that it got in. 1Password adds that an approval covers one login, not the whole vault.

The FAQ on 1Password's product page marks where that protection ends. What happens after login is up to the agent and the website, and 1Password warns that may include viewing information or taking actions in the account. Safeguards and approval frequency also vary by integration, it says.

"The old model where you hand over a credential and hope for the best doesn't work," chief product officer Aashish Tripathi said in the 1Password announcement.

Which agents, and when

The launch roster covers Meta's Muse, Anthropic's Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot and Town's Townie assistant. Timing is uneven. In its launch blog post, 1Password says Muse and Grok Bot support rolls out over the coming days, while the Claude Duet, OpenAI and Town connections are marked as coming soon. Browserbase, Browser Use, Perplexity, Ramp and five other companies are named as later partners.

1Password says Individual and Family customers on 1Password.com currently pay nothing extra, while 1Password.eu and 1Password.ca accounts are not supported.

TechCrunch, citing a same-day Axios report, says US banks now spot agents in customer accounts, and most apparently cannot tell whether customers sent them.

What we don't know yet

Whether banks and other sites that resist agents will treat a login approved through 1Password any differently.
When the Claude Duet, ChatGPT Work, Dots and Townie integrations will ship.
Whether business plans or 1Password.eu and 1Password.ca accounts will get the feature.