1Password for Agents lets AI bots sign into accounts without ever seeing passwords

1Password has introduced a feature that lets AI agents sign into online accounts on a user's behalf without the bot ever seeing or handling the actual credentials. According to The Verge, the feature is called 1Password for Agents, and it is not a blank check: every sign-in request from an agent still has to be approved by the account owner before anything happens.

How the flow works

As The Verge explains it, the sequence begins when an agent hits a login form. The agent proves its identity to 1Password using a digital key unique to it. The app then notifies the user, spelling out which account's credentials are being requested and the reason for the request. Once the user approves, 1Password completes the sign-in itself, entering the stored login details into the form rather than passing them to the agent, and the bot carries on with its task having never been exposed to the password.

All approvals happen on the user's own device, an arrangement 1Password compares to two-factor authentication in that the human remains the final gatekeeper for access.

CEO Aashish Tripathi framed the launch as a break with how credentials get shared today. "The old model where you hand over a credential and hope for the best doesn't work," he said in a press statement reported by The Verge, adding that an agent "should be able to do the job it's been asked to do without getting permanent access to everything behind the account."

Which agents work with it

At launch, 1Password for Agents supports Meta Muse, ChatGPT Work and Dots, Claude Duet, Grok Bot, and Town, according to The Verge. A further set of partners, including Browserbase, Browser Use, Dana, Instinct, Kernel, Perplexity, Ramp, Tomo, and Wajo, is building support, with 1Password saying more integrations will follow.

Why it matters

AI agents are shifting from answering questions to completing multi-step work such as booking, purchasing, and filing, and that work keeps running into login screens. The common fixes today are poor: pasting a password into an agent's prompt, sharing a session cookie, or storing credentials inside the agent's own environment. Each of those turns a single task into durable, largely unaudited access to an entire account, and to everything that account can reach.

1Password for Agents offers a brokered alternative. The vault sits between the agent and the website, keeps the secret to itself, and releases only an authenticated session, one approved request at a time. Structurally, it echoes what OAuth did for third-party apps: replacing credential sharing with scoped, delegated access, now applied to bots that drive browsers on your behalf.

Two open questions temper the promise. First, coverage: the model only helps where integrations exist, so agents outside the partner list are unaffected. Second, approval fatigue: if sign-in notifications become routine, users may start rubber-stamping them, quietly rebuilding the very hand-over-and-hope dynamic the feature is meant to eliminate. The Verge's report does not detail how approvals are scoped, per task, per site, or per session, nor how signed-in sessions get revoked, and those specifics will determine how much real protection the design delivers in practice.