Password manager 1Password announced on October 22 a new feature, “1Password for Agents,” that lets AI agents log in to web services without seeing the credentials. Once users approve, 1Password enters the credentials directly on the target website. Integrations with Meta’s “Muse” and SpaceXSI’s “Grok Bot” will begin rolling out over the next few days.
You want to delegate to AI, but login gets in the way
The number of people who have tried to delegate research or arrangements to an AI agent, only to get stuck at a login screen along the way, is sure to grow. When authentication is required, users have had to choose between entering their password themselves or handing their credentials to the AI.
The first option creates a contradiction: “I thought I’d delegated the task, but I still have to stay nearby.” The second is more troubling, because it means entrusting an important key to something whose actions are difficult to see.
This new feature is designed to bridge the gap between “I want to delegate” and “I don’t want to share my secrets.” The company says passwords and one-time authentication codes are never passed to AI agents. The AI handles the task, while you keep the key. It may be a bit like opening the front door yourself, rather than giving a spare key to a service provider.
1Password enters only the credentials you approve
The process is simple. When an AI agent reaches a login screen, it requests permission to use credentials saved in 1Password. Once the user reviews and approves the relevant login information, 1Password enters it on their behalf.
One example the company gave is asking Meta’s AI agent “Muse” to prepare for an event. Users can let it handle tasks such as creating invitations, managing RSVPs and finding needed supplies, while approving logins only to the sites it needs. The company says Muse cannot access any of the other login information stored in 1Password. The integration can be disconnected at any time.
What is noteworthy here is that the main benefit is not “autofill” itself. The key is that people can choose each time which login information to let the AI use. As AI becomes more capable, the question will shift from “What can it do?” to “How much access should it have?” The idea of granting finely scoped permissions also applies to managing accounts shared with contractors at work.
Support will roll out in stages, and availability varies by region
Support will expand in stages. Integrations with Meta’s “Muse” and SpaceXSI’s “Grok Bot” will begin rolling out over the next few days. The company says it plans to follow with integrations for Anthropic’s “Claude Duet,” OpenAI’s “ChatGPT Work” and “dots,” and Town’s “Townie,” among others.
Some headlines have said “ChatGPT support is coming,” but the product named so far is “ChatGPT Work,” and the integration is still planned, not available. The features, supported platforms and security measures are said to vary by integration, so the offerings will not be identical even when described as integrations.
As for cost, users of 1Password.com’s “Individual” and “Family” plans can use the feature at no additional charge. Accounts on 1Password.eu and 1Password.ca are not currently eligible. Even among 1Password users, availability depends on the type of account they have.
How to hand over the “keys” in the age of agents
The usefulness of AI agents is not determined solely by their ability to write or research. In real-world work, most information and actions are accessible only after logging in. How access to accounts is managed will determine whether agents can be used in everyday life.
The newly announced design puts permission-management principles into practice: inserting an approval step, limiting access to specific login information, and allowing the integration to be disconnected at any time. It will be worth watching how support expands beyond Muse and Grok Bot to ChatGPT Work, Claude Duet and others, including how each integration differs.
Comments