1Password launched its 1Password for Agents platform on Thursday. An AI agent can sign in to a user's account but never receives the password itself — the manager fills the fields directly on the destination site, after a human approves. The launch landed the same day Axios reported that US banks had begun detecting agent activity inside customer accounts.
Key takeaways
- The agent presents a unique digital key confirming its identity
- The user gets an alert about the sign-in attempt and approves it personally
- 1Password fills the credentials on the site itself — the agent never sees them
- Integrations: Claude Duet, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta Muse, Town agents
- A Chase spokesperson told Axios customers should not give their usernames and passwords to anyone
The two bad options until now
An agent reaches a sign-in page and stops. The user could then take over the session and type the details manually, which is slow and tedious, or paste the password into a prompt — handing it over for longer than one task.
Handing over a password, secret, or credential is a common shortcut to an agent, but can leave people with sensitive information in places they can no longer control as a copy gets saved for the next time.
Statement from 1Password's company blog.
The new flow separates those roles. The agent requests a credential from the vault, the user approves, and 1Password inserts it on the destination site itself.
| Option | What the agent gets | Problem |
|---|---|---|
| Human takes over the session | nothing | slow and tedious |
| Password pasted into a prompt | the full credential | a copy stays beyond one task |
| 1Password for Agents | approval for a single sign-in | needs a human to approve |
What this does not fix
Banks, according to Axios, cannot tell agent activity requested by a customer from the rest. This is not about rogue agents raiding accounts but the scale of automated action on users' behalf, finances included. Bankers interviewed by Axios called the situation horrifying.
Why it matters
A layer is appearing that the agent stack lacked: agent identity separated from human identity, with an explicit moment of consent. Without it, every delegated task means copying a secret. The catch is that the site receiving the sign-in still does not know it is an agent — with no signal on the service side, banks and shops can neither enforce nor audit it.
What's next
- 1Password says it is expanding the list of supported agents
- On the bank side there is still no signal whether an agent's action is customer-authorised
