1Password launched 1Password for Agents on Thursday, October 22, a feature that lets AI agents sign in to your online accounts without ever seeing your passwords. The agent asks for access, you approve it inside 1Password, and the password manager fills in the login itself. If you’re curious about AI assistants that do tasks on their own, it gives you an alternative to handing over a password.
How does the login work?
TechCrunch and 9to5Google describe the same four steps:
- The agent reaches a sign-in page and asks 1Password for the credential it needs.
- It proves its identity by sending 1Password a unique digital key.
- You get an alert, see which credential the agent wants, and can approve or deny it.
- If you approve, 1Password fills the details straight into the form. The agent finishes the task but never receives the password.
According to 9to5Google, once the site accepts the login, 1Password removes the credential, so the agent can’t read it or reuse it later. The rest of your vault stays out of its reach too.
Think of it like asking an employee to open a door for you instead of giving them a copy of the key. The job gets done, but the key never leaves your hand.
Why does this matter?
An AI agent is a program that doesn’t just answer questions but does things for you, like booking a trip, paying a bill or filling out a form. To do that it often needs to sign in to personal accounts. The most common shortcut is to give the agent your username and password, and 1Password argues that shortcut leaves sensitive information somewhere you lose control over it.
The problem already exists outside the lab. According to TechCrunch, citing an Axios report from the same day, U.S. banks have started detecting agent activity in customer accounts, and most can’t tell whether it reflects the customer’s own request. Bankers quoted by Axios called the situation “horrifying,” and a Chase spokesperson told customers they should not give their usernames and passwords to anyone.
Which agents work with it?
The sources don’t fully agree here. 9to5Google says Meta’s Muse and Grok Bot work today, with OpenAI’s Dots and Anthropic’s Claude Duet on the way. TechCrunch lists those agents among the ones that can use the platform, along with agents from the assistant startup Town. 1Password says it’s working with other companies to expand the list, so check the app to see what’s enabled for your account before counting on a specific agent.
What changes for you?
If you use 1Password, the idea is simple: the agent has to ask permission and you decide. That gives you an easy rule for any AI assistant from any company: never type your password into the chat with the agent. If it asks, be suspicious.
A few habits that help no matter which tool you use:
- Read the alert before approving. Check which account and which credential are being requested.
- Deny what you don’t recognize. Approving on autopilot defeats the protection.
- Keep two-step verification on important accounts like email and banking. If an account has already slipped out of your control, our step-by-step guide to recovering a Gmail password is a good place to start.
- Understand the risk of agents with broad access. In another story we covered, OpenAI models used a crypto wallet without permission, a reminder that giving an agent too much power has a price.
Is it worth waiting?
The idea answers a real question: how to get the convenience of agents without handing over the keys to everything. But the list of compatible agents is still short, and two big names show up only as “coming soon” in at least one source. If you already pay for 1Password, you can switch it on once your favorite agent is supported. If you don’t use a password manager yet, this is one more reason to start, and in that case it’s worth comparing options carefully before moving all your accounts.
1Password won’t be the only company trying to solve this, because the problem grows along with agent use. What to watch over the next few months is whether banks and big sites start accepting this kind of delegated access officially, or keep blocking any automated activity.
