How the new tool works
1Password unveiled 1Password for Agents on Thursday, a feature that allows an AI agent to access a user's personal accounts without ever taking possession of the credentials. TechCrunch reports this. According to the company, the mechanism is based on an exchange of information among the agent, the account holder, and the password manager: when the automation attempts an access, it sends 1Password a digital key certifying its identity; the user receives an alert and must confirm the request; only at that point does the service perform the login on the agent's behalf, filling in the credentials.
The feature is designed to be vendor-independent: 1Password lists among compatible agents Anthropic's Claude Duet, OpenAI's ChatGPT Work and Dots, Grok Bot, Meta's Muse, and those from the company Town, while specifying that it is working with other companies to expand the offering. In its post, the company explains that directly entrusting a password or secret to an agent is a widespread shortcut, but it leaves sensitive data in places outside the user's control, because a copy is retained for subsequent accesses.
Safeguards and limits of the model
The distinctive point is the separation between the delegated task and the secret that makes it possible. In the description provided, it is 1Password that enters the credentials at login, while the agent remains outside; the user's control is exercised at the confirmation stage. However, TechCrunch does not report technical details on how the agent's identity is verified, on procedures in the event of a compromised key, on how long access logs are retained, or on the service's availability. These are the aspects to verify in order to assess the strength of the promised protection.
The question also remains open as to what, in concrete terms, authorizing an agent means: does the confirmation concern the individual access or the entire operation that the agent performs after login? The available material does not clarify this, and this distinction affects the actual scope of the control left to the user.
The push from bank accounts
The news comes on the same day that Axios reports that U.S. banks have begun detecting agent activity within customer accounts. According to TechCrunch, this would not necessarily be malicious automation, but rather consumers' growing habit of delegating tasks, including financial ones, to automated services. The problem highlighted is that in most cases institutions cannot determine whether that activity is tied to a customer request.
TechCrunch reports that bankers interviewed by Axios described the situation as "horrifying" and that a Chase spokesperson said customers should not give their username and password to anyone. The emergence of 1Password for Agents should be read in this context: security is trying to adapt to a use of agents that institutions struggle to distinguish from ordinary access. That the proposed model solves the banks' problem is not supported by the material, nor is it known whether financial institutions have adopted or evaluated the tool.
