1Password and Anthropic have enabled AI agents, such as Claude, to perform tasks that require logging into digital platforms without needing to share passwords or verification codes with the agent itself.

This comes through a new integration between the two parties that allows agents to complete identity-verification processes securely and flexibly.

The new solution, currently available in a trial version, relies on a mechanism through which login data passes directly from 1Password to the required website without being accessible to the agent or the AI model itself.

When Claude attempts to complete a task that requires entering login credentials, a notification appears to the user explaining the required service, and the user can approve it via fingerprint.

1Password enters the information directly into the website's page, and the permission expires when the task ends, so the agent needs to re-request authorization for future times.

This integration reduces the risks of sensitive data leaks and protects against known prompt injection attacks, as the agent does not have broad or continuous access permissions and does not obtain passwords directly.

The application also ensures that no sensitive information appears after entering login data, and clears the fields if the process fails, adding an extra layer of protection.

Hackers use AI to discover an unannounced vulnerability

The idea is gaining increasing importance with the rapid growth of using AI agents in financial, commercial, and administrative tasks, as recent reports indicate that the delay in appropriate digital identity solutions costs companies around the world billions of dollars annually due to fraud or incomplete operations.

The new 1Password mechanism gives institutions and users a way to precisely control the delegation of temporary login tasks without sacrificing security, and opens the door for companies in the financial and tech sectors to benefit from AI capabilities without exposing customer data or systems to risk.

Go to top button