1Password lets AI agents sign in without seeing a password

The main way an unattended agent loses control of an account is a copy of a working password it was handed and now has to protect. 1Password announced 1Password for Agents on Thursday morning, a platform that lets a personal AI agent sign a user into a service without ever seeing or storing the credential it is using. It names the agents from Meta, Anthropic, OpenAI, SpaceXSI and Town as launch partners, with nine more companies working on integrations.

The announcement came out of Toronto at 9:00am Eastern time, and the mechanism is a key exchange. When an agent reaches a sign-in form, it uses a unique digital key to prove its identity to 1Password. The person then sees which credential the agent wants and why, on their own device, and approves or refuses it there. 1Password fills the field directly into the sign-in form on approval. The agent carries on with the signed-in session; the password never passes through it.

Which agents are integrated at launch

1Password names five agents as integrated at launch: Muse from Meta, Anthropic's Claude Duet, OpenAI's ChatGPT Work and Dots, SpaceXSI's Grok Bot, and Town's Townie. Nine other companies are working on integrations without a date attached: Browserbase, Browser Use, Dana, Instinct, KERNEL, Perplexity, Ramp, Tomo and Wajo.

The named launch partners each attached a quote to the press release. "Claude Duet asks before it signs in anywhere, and soon 1Password will complete the sign-in with the passwords people have already set up," said Ami Vora, Chief Product Officer at Anthropic. Glen Coates, Head of App Platform at OpenAI, said the integration gives agents a way to use "accounts users already have saved in 1Password without exposing sensitive credentials to the model." Nat Friedman, Head of Product at Meta Superintelligence Labs, named himself as a 1Password customer since 2007.

What the trust layer is replacing

"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. The old model is the one every agent still uses by default. A user pastes a password into a chat so an agent can log in, or sets up a session the agent inherits. Either way, a copy of the credential now lives in a place the user no longer controls, and revoking it means changing the password everywhere else that uses it.

1Password's own figures state the scale of what the system plugs into: more than 1.6 billion credentials and secrets held in its enterprise vault, used by more than 200,000 businesses the company names, including Canva, Cursor, Figma, GitHub, Hugging Face, MongoDB, Notion, Perplexity, Salesforce, Stripe, Vercel, Wiz and Workday.

A team evaluating one of the named agents can now ask a different question than before: not "do we trust the agent with our passwords," but "do we trust the agent with the approval prompts." The 1Password press page lists availability of integrations and documentation on its site and blog but gives no public date for general availability. A team wanting the shape of this control before adopting an agent should test the approval flow on one connector first and watch what the agent asks for.

Source

1Password: 1Password for Agents launches as the trust layer for personal AI agents. Press coverage: The Verge: 1Password lets AI agents use your passwords without seeing them, by Jess Weatherbed.