1Password now allows AI agents to log into your accounts without knowing your passwords

, ,

1Password, a password manager publisher, announces 1Password for Agents, a platform that allows AI agents to access online accounts necessary for their tasks without providing them with the credentials. Announced on October 22, it targets a sensitive step in these uses: authentication.

  • 1Password permet aux agents IA de se connecter à des comptes sans connaître les mots de passe.
  • L’utilisateur autorise chaque demande de connexion depuis son propre appareil.
  • Le gestionnaire remplit lui-même les champs, sans transmettre les identifiants au modèle d’IA.
  • Le service est gratuit pour les abonnés Individual et Family, mais pas pour les comptes 1Password.eu et 1Password.ca.

Beyond text generation or data analysis, agents—AI programs capable of acting on behalf of a user—are now expected to intervene on online services, prepare expense reports, or manage administrative tasks. To perform these operations, they often need to log into services, which requires authentication.

Authentication without transmitting passwords

Until now, assigning a task to an agent could involve sharing a password with it, for example in an instruction, or taking over oneself to log in. The first method exposes credentials to risks of storage or reuse, while the second limits the benefit of automation.

With 1Password for Agents, the agent makes an access request when it encounters a login page. It identifies itself to the manager using a unique digital key, then the user is invited to review the request and authorize it from their own device. The manager then directly fills in the fields of the target site, without transmitting the password to the AI model.

The principle is to separate the authorization to use an account from the knowledge of the associated credentials. The agent can continue its work once logged in, while the secrets remain in the secure storage space. According to the publisher, the mechanism also targets one-time codes.

Integrations still being rolled out

The service is expected to work with several agents, including Muse from Meta, Grok Bot, Claude Duet, ChatGPT Work and dots, as well as Townie, Town’s assistant. The schedule varies according to partners: some integrations are being deployed, others are announced for a later stage. Browserbase, Perplexity, and Ramp are also among the players involved in the planned developments.

At launch, the offer is provided at no extra cost to Individual and Family subscribers with an account on 1Password.com. Deployment remains subject to certain geographic restrictions: accounts hosted on 1Password.eu and 1Password.ca are not yet eligible. Access terms and available features also vary depending on the agent and its level of integration with the credential manager.

Access control remains crucial

The initiative illustrates an evolution in identity management: agents become intermediaries capable of acting on behalf of a user, which requires better regulation of their rights. Protecting the password alone is not enough to guarantee that an operation will be legitimate or that the agent will limit itself to the requested task.

1Password’s system thus relies on human approval and on the separation between the agent and authentication data. Its effectiveness will also depend on the robustness of integrations, the traceability of operations, and the possibilities for revoking access.