1Password’s new access layer allows AI assistants to use stored credentials only after user approval, as banks and security experts warn about the risks of autonomous account access.
1Password has unveiled a new access layer designed for a fast-emerging problem in consumer and workplace AI: how to let software act on a user’s behalf without handing over the password itself. The company’s new 1Password for Agents system is intended to let AI assistants sign in to services while keeping credentials inside 1Password, rather than exposing them to the model or storing them in the agent’s own context. According to 1Password’s announcement, the first integration is with Meta’s Muse, with support also planned for ChatGPT Work, OpenAI’s Dots, Anthropic’s Claude, Grok Bot and other agent platforms. The user remains in control, approving or rejecting each request before access is granted.
The process is deliberately narrow. When an agent reaches a login screen, it requests access from 1Password, which then prompts the user to decide whether that specific account should be unlocked. If approved, 1Password inserts the credentials directly into the form and keeps the agent from seeing the password. The company says the agent cannot approve its own request, a safeguard that matters because the entire purpose of the system is to separate identity from automation. In practical terms, that could allow an assistant to book travel, shop online or sign into work tools without the clumsy cycle of pasting passwords into a chat window or repeatedly taking manual control of the browser.
The move reflects a broader shift in the security debate around AI agents. As these systems become more useful, they are being given access to emails, calendars, shopping accounts, financial services and business applications. That creates a harder question than simply whether an agent is intelligent enough to complete a task: it is whether anyone can reliably tell when it is acting for the user and when it is acting on its own. TechCrunch reported that the problem is already surfacing in bank accounts, where institutions are struggling to distinguish authorised automation from activity that may breach policy or create liability.
That concern is increasingly visible across the financial sector. Reports from Tom’s Hardware and Axios say several major US banks, including Morgan Stanley, Bank of America, JPMorgan Chase and Wells Fargo, have moved to restrict third-party AI agents because some have been detected trying to access customer accounts or even initiate transactions. The wider security context is also worsening. Microsoft has warned that frontier AI systems cannot simply be trusted to follow organisational rules, highlighting risks from unpredictable behaviour, cross-prompt injection and coordinated misuse. Against that backdrop, 1Password is positioning itself not as the maker of a single winning agent, but as a trust layer that could sit between users and whichever assistants they choose to use. In that sense, the company is making a clear bet: the next important platform in AI may not be the most capable model, but the one that can prove it only gets permission, not the password.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
