1Password Wants AI Agents to Work Without Ever Holding Your Password
1Password Wants AI Agents to Work Without Ever Holding Your Password
1Password has launched 1Password for Agents, a system designed to let AI assistants log into personal accounts while keeping the underlying username and password out of the agent’s hands. The move lands as consumers increasingly ask automated tools to handle real-world tasks — including potentially sensitive financial ones.
The process is deliberately gated. When an agent reaches a sign-in page, it presents a unique digital key to verify its identity with 1Password. The user then receives an alert identifying the account and the reason for the request. Only after approval on the user’s own device does 1Password fill in the credentials and allow the agent to continue.
For 1Password, that is a sharp break from the familiar — and risky — shortcut of simply giving a bot a password. “The old model where you hand over a credential and hope for the best doesn’t work,” CEO Aashish Tripathi said, arguing that agents should complete assigned work without gaining permanent access to everything behind an account. The company says the service can work with agents including Claude Duet, ChatGPT Work and Dots, Grok Bot, Meta Muse and Town, with further integrations planned.
But the launch also exposes the broader trust problem. On the same day, U.S. banks were reported to be detecting agentic activity inside customer accounts, even when they could not reliably tell whether it had been authorized by the account holder. Bankers described that uncertainty as “horrifying,” while a Chase spokesperson warned customers not to share usernames and passwords with anyone.
1Password’s answer is not to stop agents, but to put every consequential login behind a fresh, user-controlled approval. Whether that becomes a workable security layer — or merely adds friction to a fast-moving automation economy — will depend on how reliably it can prove who, exactly, is asking to get in.
Write a comment