AivexaNewsSearch
AI news for builders and product teamsChecked every hour

1Password launches a way for AI agents to log in on your behalf without seeing your password

Collected Oct 22, 2026

1Password announced on Thursday a product called 1Password for Agents, which lets an AI agent log into a customer's personal accounts on their behalf while keeping the underlying credentials out of the agent's hands. The company framed the release as a way to replace a common but risky shortcut — passing a password directly to an assistant — with a mediated flow it controls.

The mechanism has three parties: the agent, the user, and the password manager. When an agent tries to sign in for a user, it sends a unique digital key to 1Password that confirms the agent's digital identity. The user receives an alert about the attempt and can personally approve or decline it. If approved, 1Password performs the sign-in and fills in the credentials itself, letting the agent continue with the task without ever holding the secret.

That design matters because of what typically happens today. As 1Password's blog post put it, handing a password or secret to an agent often means a copy gets stored somewhere the user no longer controls, ready for reuse next time. Under the new model, the credential stays inside the password manager's boundary and the agent only receives proof that authentication succeeded.

1Password says the service is not tied to a single assistant. It lists support for Claude Duet from Anthropic, OpenAI's Dots and ChatGPT Work, Grok Bot, Meta's Muse, and agents from the AI assistant startup Town, and said it is working with additional companies to broaden that list.

The timing is pointed. The launch came the same day Axios reported that U.S. banks have started detecting agentic activity inside customer accounts. According to that reporting, the activity is largely not rogue agents raiding accounts but consumers letting automated services act for them, including on financial matters. The core problem bankers described is that most banks cannot determine whether a given automated action was authorized by the user. Bankers quoted by Axios called the situation "horrifying," and a Chase spokesperson told the outlet that customers "should not give their usernames and passwords to anyone."

This is the gap 1Password appears to be targeting: the difference between an agent acting with the user's consent and an institution being able to tell that consent existed. On the 1Password flow, the approval step is what separates the two, because the user is notified at the moment of each login attempt and confirms it. In practice, that turns a static secret into a per-request authorization, which is closer to how enterprises treat machine access internally — short-lived, explicitly granted, and revocable — than to how consumers usually handle passwords.

The trade-off is friction and coverage. A confirmation alert for every sign-in could become tedious for agents that log into many services, and it only protects accounts whose credentials live in 1Password and whose sites cooperate with automated filling. Whether bank or other high-risk sites will accept the resulting login sessions, and how they distinguish it from the activity they already find hard to classify, is an open question. This suggests the product is likely most useful as a consumer-side control layer, not a fix for how institutions verify agent activity on their own end.

Why it matters: developers building agent products now have a vendor-supported path to let agents complete authenticated tasks without handling raw passwords, which may become an expected integration rather than an extra. Security and product teams should expect users to ask for credential delegation that looks like this — user-visible approval, credentials never exposed to the model — and should plan for the friction that per-login confirmation introduces. In my reading, the more consequential effect is on trust: institutions that currently cannot tell authorized agent behavior from the rest may eventually demand exactly this kind of verifiable consent, and products without it could face resistance from both users and the services they touch.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

The new feature allows agents to log into a customer's personal accounts while still protecting the private credentials that secure those accounts.