AI Agents Finally Get Access Without the Password

AI Agents Finally Get Access Without the Password

Given free rein, AI agents can research products, compare prices and fill out forms. They’re thwarted the moment a task requires a login. Until now, the only options have been handing over the password and accepting the security risk, or taking over the keyboard every time a sign-in screen appears, defeating the point of having an agent in the first place. Now, 1Password and Anthropic have launched an integration that removes that bottleneck.

1Password for Claude, currently in beta, lets Claude complete tasks that require usernames, passwords and authentication codes without those credentials ever reaching the AI model, 1Password wrote in a Thursday (July 16) blog post. When Claude hits a login screen during a task, it asks 1Password for the credential. A prompt appears showing the user exactly which credential is being requested and why. The user approves with a fingerprint. 1Password then fills the password directly into the website. Claude never sees it. Access ends when the task ends. The next time Claude needs the same login, it has to ask again, Help Net Security reported.

Claude Gets Access Without Ever Seeing the Password

That distinction is the architecture. A user approving a fingerprint prompt is telling the system: use this login for this task. They are not handing the AI anything it can read, store or reuse. “The answer isn’t handing agents your secrets,” Nancy Wang, CTO of 1Password, said in the release. “It is to let a user give an agent permission to use a credential without letting the agent see it.”

After the login is filled, 1Password checks that nothing sensitive was exposed on the page. If the sign-in fails, it clears what was entered before returning control to the agent.

The timing matters. Security researchers recently showed that artificial intelligence browser tools can be tricked into leaking user credentials through prompt injection attacks, hidden instructions embedded in web pages, with Claude’s own browser extension among those affected, The Next Web reported. The 1Password integration is a direct response. When an AI agent takes control of the browser, the 1Password extension automatically restricts vault access to only the credentials approved for that specific task, 9to5Mac reported. Everything else stays locked.

Outdated Identity Controls Cost Businesses Nearly $100 Billion a Year

The broader stakes are commercial. AI agents are already booking travel, managing accounts and processing procurement requests at speeds no human oversight can match, PYMNTS reported. The infrastructure to verify an agent’s identity, its authorization and whether it is operating within approved limits does not yet exist at scale. Outdated identity controls are costing businesses nearly $100 billion annually in fraud losses and missed business, according to PYMNTS Intelligence research conducted with Trulioo.

The 1Password model points toward one answer. Rather than giving an AI agent standing access to accounts, it gives the agent time-limited, task-specific permission that expires automatically. “Authorize once and trust indefinitely is not a security model; it’s a liability,” CrowdStrike CTO Elia Zaitsev said in announcing the company’s new agent identity platform, as reported by PYMNTS.

For banks, retailers and payment platforms deploying agents to handle customer-facing tasks, the question of how an agent authenticates, and what happens when it cannot, is no longer theoretical.