1Password for Agents lets an AI assistant sign in to your accounts without ever seeing your password. The assistant asks, you approve on your own device, and 1Password types the credential into the login form for it. It launched on October 22, 2026, and it tackles one of the riskiest habits in the agent era: pasting passwords into a chatbot.
Here are the facts worth knowing before you try it.
- Announced: October 22, 2026, by 1Password.
- Works with Meta’s Muse, Anthropic’s Claude Duet, OpenAI’s ChatGPT Work and dots, Grok Bot, and Town.
- You approve every sign-in on your own device.
- Pricing and a general availability date have not been published.
How 1Password for Agents keeps your password away from the AI
When an agent reaches a sign-in form, it first proves its identity to 1Password with a unique digital key. You then see which credential it wants and why, and you approve or decline. If you approve, 1Password fills the form directly. The password never travels to the agent.
The agent gets task-scoped access, not a permanent key to the whole account. It finishes the job, and the password stays in your vault, according to 1Password’s launch announcement.
The table below shows how that compares with the old habit of handing an agent your login.
| What matters | Pasting a password into an agent | 1Password for Agents |
|---|---|---|
| Can the agent see the password? | Yes, and it may keep a copy | No, 1Password fills the form itself |
| Who approves a sign-in? | Nobody, once you paste it | You, on your own device, each time |
| How much access does the agent get? | Whatever the password unlocks | Task-scoped access |
| Where does the password live? | In the chat or agent memory | In your 1Password vault |
The point is simple: an agent can only leak or misuse what it never held.
Which AI agents work with it today
At launch, five agents are named in the TechCrunch report and 1Password’s own announcement.
- Muse from Meta
- Claude Duet from Anthropic
- ChatGPT Work and dots from OpenAI
- Grok Bot
- Town, from the startup Townie
Don’t expect everything to work on day one. 1Password’s page says an Anthropic executive expects sign-ins for Claude Duet to be completed “soon,” so some integrations are still being finished. Partners such as Browserbase, Browser Use, and Perplexity are also building on the platform.
Why banks and security teams care about agent logins
The launch landed as U.S. banks reported spotting agent activity in customer accounts and struggling to tell whether it reflects what the customer asked for. A Chase spokesperson told Axios that customers “should not give their usernames and passwords to anyone,” as relayed by TechCrunch.
We covered a related warning in our post on the Bank of America AI agent password warning. Together, the two stories point the same way: sharing a raw password with software is the weak link.
Old agent logins are a problem too. Our piece on dormant AI agents that still hold live credentials shows why short-lived, approved access beats a password that never expires.
What 1Password still has not said
The announcement is light on practical details, so hold off on rebuilding your routine around it.
- Pricing: neither the press release nor TechCrunch lists plans or costs.
- General availability dates for each agent are not given.
- Supported browsers are not listed.
Those gaps matter because a security feature is only useful if you can turn it on today. Check the integrations page on 1Password’s site before you plan around it.
Safer habits while you wait for wider support
You don’t need the new feature to cut your risk today. Start by never typing a password into an agent’s chat box, even if the agent asks politely. If a task needs a login, ask yourself whether it really needs your main account.
Where an agent must act for you, use a separate account with limited rights, and turn on two-factor sign-in for it. Then review which agents still have access every few weeks and remove the ones you no longer use. It takes five minutes and shrinks the damage if one of them is ever compromised.
Frequently Asked Questions
Does the AI agent ever see my password?
No. According to 1Password, the credential goes straight from your vault into the sign-in form, and the agent never receives or stores it.
Do I have to approve every sign-in?
Yes. Each login asks for your approval on your own device, and you can see which credential is requested and why.
Is 1Password for Agents free?
1Password has not published pricing in its launch announcement. Check your plan’s details before assuming it is included.
Which AI assistants can use it?
Launch partners are Muse, Claude Duet, ChatGPT Work, dots, Grok Bot, and Town. More are planned.
Should you use it?
If you already let an agent shop, book, or manage accounts for you, this is the safer way to do it, and it beats pasting a password into a chat window. But wait for clear pricing and confirmed support for your specific agent before switching. Until then, keep sensitive accounts like banking out of any agent’s reach.
Leave a Reply