1Password lets AI agents connect without seeing passwords

1Password launches “1Password for Agents,” a system that allows AI agents to connect to services without seeing passwords. Access remains subject to your approval, on your device.

1Password has introduced “1Password for Agents,” a system that allows AI agents to connect to accounts without accessing the passwords themselves. The password manager aims to solve a central problem with assistants capable of performing online tasks: how to give them useful access without opening the entire vault. User approval is required for each connection request, and the operation takes place on the user's own device.

Un agent se connecte, mais la clef reste dans la poche

The principle is based on a digital identity unique to the agent. When it reaches a login form, it authenticates with 1Password using this key, then the service shows the user which credential is requested and why. If access is approved, 1Password fills in the login fields itself; the agent never obtains the password. Once the session is open, the information is removed so it cannot be read or reused.

The mechanism is conceptually similar to two-factor authentication: the user retains control over validation, and the agent operates only after explicit consent. 1Password also emphasizes a very concrete governance point: the rest of the vault is not exposed to the agent. This is the difference between “helping to do the task” and “giving the keys to the truck.”

Quels agents sont compatibles, et jusqu’où va le support ?

According to 1Password, the system is designed to work with several popular agents, including Meta Muse, ChatGPT Work, Dots, Claude Duet, Grok Bot, and Town. The company also mentions integrations in preparation with Browserbase, Browser Use, Dana, Instinct, Kernel, Perplexity, Ramp, Tomo, and Wajo, with more connections to come. 9to5Google indicates, however, that at this stage support is limited to Muse and Grok Bot, with other arrivals still announced [to be verified].

This caution is not trivial. The AI agent market is moving fast, but real-world uses quickly face the issue of credentials, especially when navigating third-party services. By offering an intermediate layer, 1Password positions itself on very concrete ground: operational trust, not talk about the “magic” of agents.

Pourquoi ce lancement compte pour les usages professionnels

In professional environments, an AI agent often needs to go through existing tools: CRM, messaging, cloud services, banking interfaces, or business software. Sharing a password with an agent remains a bad idea, as the agent can store, transmit, or reuse it. 1Password therefore seeks to provide a compromise: let the agent act, but without granting permanent access to accounts.

The signal sent to the market is clear. Agents are no longer just demonstrators capable of writing text or summarizing documents; they are beginning to perform actions on real services. And as soon as the action involves a login, security becomes the main issue again. Without this, automation remains somewhat theoretical, and sometimes frankly stressful.

By the numbers

  • 2 — two-step control, with user approval and filling by 1Password.
  • 6 — agents cited as compatible or announced in 1Password's presentation.
  • 10+ — integration partners mentioned by 1Password for future support.
  • October 22, 2026 — date of publication of the first announcements reported by The Verge and 9to5Google.

To read