1Password partners with Meta and OpenAI to let AI agents log in via digital keys while users control credential access
The new 1Password for Agents feature lets AI assistants log into accounts via digital keys, keeping passwords hidden until users approve each request.
Key points
- Password launched a platform allowing AI agents to sign in without accessing raw passwords.
- Users must approve each login request on their device before credentials are filled.
- Initial partners include Meta, OpenAI, Anthropic, and SpaceXSI, with more integrations planned.
Identity security firm 1Password launched 1Password for Agents, a new system that allows personal AI agents to log into user accounts without ever seeing or handling the actual passwords. According to The Verge, the feature addresses security concerns as AI agents begin performing tasks like booking travel and paying bills, which previously required users to paste credentials into chat windows where copies could remain outside their control.
The platform uses a digital key to verify the identity of the AI agent when it encounters a login form. Instead of handing over the password, 1Password sends a notification to the user’s device detailing which account is needed and why. The user must approve the request on their own device, similar to two-factor authentication, before the application fills in the login information directly into the web page.
How the Approval Process Works
Aashish Tripathi, 1Password’s chief product officer, stated that the old model of handing over credentials no longer works as agents perform real work on behalf of users. He noted that an agent should complete a specific task without gaining permanent access to the entire account. By keeping the password in the user’s vault and only typing it into the form upon approval, 1Password ensures the agent never possesses the credential itself.
This process requires the user to remain in control of permissions. The Next Web reported that the system prevents agents from having free reign over connected accounts. Each sign-in request is treated as a discrete event that requires explicit consent, ensuring that the trust layer remains with the human user rather than the automated bot.
Partner Integrations and Availability
1Password announced several launch partners, including Meta’s Muse, OpenAI’s ChatGPT Work and Dots, Anthropic’s Claude Duet, SpaceXSI’s Grok Bot, and Town’s Townie assistant. SiliconANGLE reported that Meta’s Muse requests logins from 1Password whenever the integration is active. OpenAI has connected its ChatGPT Work and Dots agents, while Anthropic’s Claude Duet will support the feature soon, according to Anthropic Chief Product Officer Ami Vora.
Beyond the initial group, nine more companies are building integrations, including Perplexity AI, Browserbase, Browser Use, Dana, Instinct, Kernel, Ramp, Tomo, and Wajo. The Next Web noted that Browserbase and Perplexity have connected their browser-based agents to 1Password since late 2025. 1Password did not provide specific pricing or launch dates for individual partners.
What should you do?
Security professionals should monitor how their organization’s AI agents handle authentication. Users of personal AI agents should review which services are connected to 1Password and ensure they understand the approval workflow. Organizations should assess whether this model fits their security policies for automated tasks.
What to do and how to stay safe: 1Password
- Review which AI agents have access to your 1Password vault and verify their identities.
- Enable the approval requirement for all agent sign-in requests to maintain control.
- Monitor your device for unexpected login notifications from unfamiliar agents.
- Wait for vendor updates before integrating new AI tools with sensitive accounts.
Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How does 1Password for Agents protect passwords?
The system uses a digital key to identify the agent, but the password is never sent to the agent. Instead, 1Password fills the login form directly after the user approves the request on their device.
Which AI agents support 1Password for Agents at launch?
Launch partners include Meta Muse, OpenAI ChatGPT Work and Dots, Anthropic Claude Duet, SpaceXSI Grok Bot, and Town’s Townie assistant, with more integrations planned.
Is the password ever visible to the AI agent?
No, the password stays in the user’s vault. The agent verifies its identity with a digital key, but 1Password types the credentials into the page without handing them to the agent.
