1Password for Agents lets AI log in without ever seeing your password

What's going on

Password manager 1Password unveiled a new tool called "1Password for Agents" on Thursday, October 22, 2026. It lets users have AI agents (programs that carry out tasks on the web on their own) log in to their personal accounts without the agent ever receiving the credentials. The backdrop: people are giving AI assistants more and more access to their digital lives, and security vendors are catching up.

How it works

The system sets up a three-way exchange between the agent, the user and the password manager:

1. Identity: When an agent wants to sign in, it sends a unique digital key to 1Password to prove who it is.
2. Approval: The user gets a notification on their own device. According to the company, it shows which credentials are needed and why. The user approves the request in person.
3. Login: 1Password enters the credentials directly into the login form. They are then removed, so the agent can't read or reuse them.

Access applies only to the current session. It doesn't carry over to future tasks. The company argues that the common shortcut of simply handing an agent a password leaves sensitive data in places you no longer control, because a copy gets stored for next time.

"An agent should be able to get its job done without getting permanent access to everything behind the account." (Aashish Tripathi, Chief Product Officer at 1Password, paraphrased)

Supported agents

The names mentioned are Anthropic's Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta's Muse, and agents from the startup Town. The sources partly disagree on the current status. According to the aggregator NewsCord, the integration is limited to Muse and Grok Bot at launch, with support for Claude Duet, ChatGPT Work, Dots and Town's Townie planned "soon." TechCrunch lists all of them as selectable and says 1Password is working with more companies on expanding support. It's therefore not entirely clear which version is right. The two-agent limit at launch comes directly from 1Password's statements as reported by NewsCord. The sources don't mention pricing or availability.

The backdrop: banks sound the alarm

At the same time, Axios reported that US banks have started to detect agentic activity in customer accounts. This isn't about hijacked accounts. Customers are voluntarily giving automated services access, including for financial tasks. The problem: most banks can't tell whether an agent is acting on the customer's behalf or not. Bankers surveyed called the situation "horrifying." A Chase spokesperson said customers "should not give their usernames and passwords to anyone."

Our take

The approach resembles the principle of delegated access: the agent gets only time-limited access, and a human signs off on every login. That lowers the risk of passwords ending up with agent providers. What remains open is what an agent can do inside the session once it's logged in, since a signed-in session is still powerful. According to the sources, the product also doesn't directly solve the banks' problem of telling agents apart from users. No independent security audits are mentioned; the security claims come from the vendor.