1Password launches tool for AI agents to log in without seeing passwords

The rise of agentic activity

Consumers are increasingly entrusting AI agents with the ability to perform actions on their behalf. This shift toward automation means that software can now navigate websites and manage personal accounts to complete specific tasks. While this offers convenience, it creates a significant security dilemma regarding how these agents access private data.

Security services are now racing to keep up with this trend as the technology becomes more common. Banks in the United States have already begun to detect agentic activity occurring inside user accounts. This activity is not necessarily malicious, but rather a reflection of how people are using automated services for financial matters.

The primary concern for institutions is the lack of clarity regarding these actions. Bankers have noted that it is often impossible to tell if an automated action was initiated by a human request or something else. This ambiguity makes it difficult for traditional security systems to distinguish between a user's intent and a potential security breach.

How the new tool works

1Password launched a new platform called 1Password for Agents to solve this specific problem. The tool allows users to let an AI agent log into an account without the agent ever seeing the actual password. It creates a communication system between the user, the password manager, and the AI agent to facilitate these logins.

When an agent tries to log in for a user, it sends a unique digital key to 1Password to confirm its identity. The user then receives a notification about the specific login attempt and must personally confirm the request. This ensures that the human remains in control of the authorization process at all times.

Once the user approves the request, 1Password signs the agent in by filling in the credentials on the assistant's behalf. This allows the agent to complete its assigned task without the user having to share a permanent secret. By using this method, the password remains stored securely in the manager rather than being saved as a copy in the agent's memory.

Supported AI assistants

The new platform is designed to be compatible with a variety of different AI agents. Users can choose from several well-known models and assistants, including Claude Duet from Anthropic and OpenAI’s Dots or ChatGPT Work. It also supports Grok Bot and Meta’s Muse.

In addition to these major tech companies, the tool supports agents from the AI assistant startup Town. 1Password stated that they are currently working with other companies to expand the list of supported services. This suggests a goal of making the tool a standard for any automated assistant a user might choose to employ.

The goal is to provide a trusted way to delegate tasks to personal AI agents. By offering a way to manage these interactions, 1Password aims to prevent users from taking common shortcuts that leave sensitive information in places they can no longer control. This approach keeps the user's credentials private while still enabling the benefits of automation.

Security risks of shortcuts

Many users currently find it easier to simply hand over a password or secret to an AI agent to get a task done. However, 1Password warns that this is a common shortcut that can lead to significant security issues. When a password is shared with an agent, it can be saved as a copy for future use, making it much harder for the user to manage or delete.

This risk is particularly high for sensitive information like banking credentials. As reported by Axios, some bank spokespeople have warned that customers should not give their usernames and passwords to anyone. The lack of control over where a copy of a password ends up is a primary concern for security experts and financial institutions alike.

By creating a system where the agent never sees the credential, 1Password for Agents attempts to eliminate this risk. It provides a way for the agent to act as a proxy rather than a holder of the secret. This allows for a higher level of security while still allowing the automation to function as intended by the user.

The future of automation

The launch of this tool comes at a time when the industry is grappling with how to handle automated interactions. As more people use agents for complex tasks, the need for secure